Modern Creator
Open Residency · YouTube

Creator Masterclass: The Future of Human Creativity in 2027

An AI creative founder argues that taste, editing and traditional craft decide who wins with generative AI, then builds a 15-second shoe commercial live from reference frames to finished Seedance video.

Posted
3 days ago
Duration
Format
Interview
educational
Views
43.3K
928 likes
Big Idea

The argument in one line.

Generative AI rewards people with traditional craft and taste, because a convincing AI video is built from human-designed ingredients (references, script, character, product and location sheets) long before any video is generated.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • A video editor, photographer or designer wondering whether AI makes your skill worthless or more valuable.
  • A brand or marketing lead deciding where AI belongs in your creative process without alienating your audience.
  • Someone who has tried AI video tools, got generic results, and wants a concrete pre-production workflow for consistent characters and products.
  • A team lead looking for a simple, repeatable way to keep staff current on fast-moving AI tools.
  • An e-commerce operator curious what it actually takes, in steps and dollars, to make a product spot with AI.
SKIP IF…
  • You want a technical tutorial on a specific video model's settings; the demo is a workflow overview, not a parameter deep dive.
  • You're looking for a skeptical or critical take on AI in creative work; the guest runs an AI creative education business and is firmly pro-AI.
  • You only want the build; the first 30 minutes are interview and opinion before the live demo starts.
TL;DR

The full version, fast.

The interview's thesis is that AI rewards human craft rather than replacing it. Brands win by using AI where audiences won't feel cheated, often in pre-production or hybrid VFX, and the guest's company works at roughly 80% human input to 20% AI. Editors who can also generate footage become de facto creative directors, and taste, built from lived experience, is the differentiator. The live demo shows the method: collect references (ShotDeck, Cosmos, Milanote), handwrite a shot-by-shot script, use Claude to write prompts, build character, product and location sheets in OpenArt, record a phonetic voice sample, then generate in Seedance 2.5 and stitch the best seconds from several generations in an editor.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →
Voices

Who's talking.

00:45guestRourke Heath
00:00hostMark Mastrandrea
Chapters

Where the time goes.

00:00 – 01:13

01 · Trailer and intro

Cold-open clips, the 'editors will be the most expensive hire' tease, and the promise of a live A-to-Z demo at the end.

01:13 – 02:57

02 · Who wins the 78 million AI jobs

Three stats (78M net jobs by 2030, 90% of orgs expecting AI skill shortages, up to 56% higher wages) and the claim that most companies use AI the wrong way, for cost cutting.

02:57 – 06:52

03 · The North Face and the three ways to use AI

Why an athlete-driven brand keeps AI off camera, and the spectrum from full gen AI to hybrid VFX to pre-production only, illustrated with a zombie baseball-bat shoot.

06:52 – 09:25

04 · Editors and ideas people

Video editors as the new creative directors, six-figure editor salaries, and why traditional creatives beat AI-native beginners.

09:25 – 14:11

05 · Taste and where a young creative should start

Taste comes from lived experience and can't be taught; start with one traditional skill and keep AI to about 20% of the work.

14:11 – 17:23

06 · Upskilling a team

The 40-minute weekly tinkering rule from Google Flow, the creative technologist role, and AI-literate people on set at MrBeast scale.

17:23 – 21:57

07 · The copycat problem

Consuming the whole market to stay fresh, and the creator who copied the guest's scripts and visuals all the way to a Google Pixel sponsorship.

21:57 – 26:00

08 · Remixing viral videos and letting fans design

Swapping a brand mascot into an already-viral clip, and a Vans concept where fans design with AI, vote, and earn royalties.

26:00 – 30:15

09 · Jake Paul, the World Cup and brand risk

Licensing a likeness to Sora 2 for billions of views, AI World Cup fan clips, when AI can hurt a brand, and an AppLovin read.

30:15 – 33:05

10 · The finished commercial and finding references

The Tamagotchi short plays, then step one: references, avoiding Pinterest, and shooting your own street as a location reference.

33:05 – 43:20

11 · ShotDeck, Milanote and casting with Claude

Using ShotDeck frames as a competitive feed and camera-data source, mood boarding in Milanote, casting characters via Claude-written prompts, and the main image models on OpenArt.

43:20 – 49:50

12 · Handwriting the script

A shot-by-shot script that specifies shot type, camera position, duration, ambient audio and how each line is delivered emotionally.

49:50 – 57:00

13 · Live character sheet of the host

Phone reference photos, a Claude-written prompt, and a side-by-side where GPT image matches the host and Google's model misses.

57:00 – 1:03:20

14 · Product sheet for the shoe

Building a multi-angle sheet of the host's sneakers, why small labels morph, and why every object in a scene needs a sheet.

1:03:20 – 1:11:20

15 · Location reference and iterating

A cabin-in-the-woods location built from a ShotDeck frame's camera data, and why regenerating beats editing an AI image.

1:11:20 – 1:13:20

16 · Voice reference

Recording about eight seconds of a phonetically complete sentence in a flat tone so the video model can match a real voice.

1:13:20 – 1:22:10

17 · Writing and running the Seedance prompt

Three beats in 15 seconds, a timecoded prompt from a custom Claude skill, a locked global camera look, and tagged references in Seedance 2.5.

1:22:10 – 1:26:20

18 · Results, mining and recap

Two generations reviewed, hallucinated details spotted, mining the best seconds from each, and the full workflow summarized.

1:26:20 – 1:27:36

19 · Sponsor: Wispr Flow

Host demo of voice-to-text email reply.

1:27:36 – 1:41:46

20 · Quickstrike questions

Rapid-fire: AI video's real cost, three days at MrBeast, why Instagram beats TikTok and YouTube, biggest paydays, podcasts as AI-proof, and GenHQ's hiring marketplace.

Atomic Insights

Lines worth screenshotting.

  • Companies that use AI mainly to cut costs and ship faster end up alienating audiences who can see when effort was removed from creative work.
  • An outdoor brand built on real athletes climbing real mountains should keep AI off camera and use it for shot lists, mood boards and pre-production instead.
  • There are three ways to use AI in video: fully generated, hybrid (real footage augmented with AI) and fully traditional with AI only in pre-production.
  • Editors who can both cut and generate footage now control the whole story, which is why they're being called the new creative directors.
  • Hiring a strong video editor in the AI space already costs six figures, and the guest expects editors to be among the most expensive people on payroll within five years.
  • Traditional creatives given Midjourney produce work roughly ten times better than AI-native beginners because they already know what makes a good image.
  • The safe ratio for AI creative work has shrunk from 50/50 to about 80% human input and 20% AI as audiences get better at spotting slop.
  • A 25-year-old creative should learn one traditional craft first, because AI is the supercharger, not the replacement.
  • Google's Flow team keeps up with AI by giving staff 40 minutes of goal-directed tinkering time every week.
  • A dedicated creative technologist who only tests new tools and filters hype from substance can keep an entire creative team current.
  • Replacing the character in an already-viral video with a brand mascot borrows proven attention, which is about the cheapest marketing available.
  • Letting fans design products with AI, voting on the best ideas and paying royalties to winners turns an audience into a marketing department.
  • Brands with a lot of reputation to lose should be cautious with AI, while startups with nothing to lose can push it hard.
  • Most of the work in a realistic AI video happens before video generation: references, script, and sheets for every character, product and location.
  • A character sheet with front, side, back, top-down and detail shots is what keeps a person looking the same when the camera cuts to a new angle.
  • The location reference locks in the color grade of an AI video, so it deserves as much care as the characters.
  • Regenerating from an updated prompt beats editing an AI image, because every AI edit degrades the picture further.
  • Roughly eight seconds of someone reading a phonetically complete sentence in a flat tone is enough voice reference for a video model to match their voice.
  • A 30-second generation on a top video model costs around $30, and a full project can take 200 to 300 generations, so AI video is neither cheap nor easy.
  • Professional AI video is made by mining the best two or three seconds from several generations and stitching them together in a normal editor.
Takeaway

AI video quality is decided before you hit generate

WHAT TO LEARN

The people winning with generative AI bring traditional craft and taste, keep AI to a minority of the work, and design every character, product and location before generating a single frame.

02Who wins the 78 million AI jobs
  • Most companies use AI to cut costs and speed up delivery, which annoys audiences who can feel when effort was removed from creative work.
  • The World Economic Forum projects a net 78 million new jobs from AI by 2030, with the gains going to people who reskill.
03The North Face and the three ways to use AI
  • Decide where AI fits by asking what your audience values; a brand built on real athletes should use AI behind the scenes, not on camera.
  • Choose deliberately between full gen AI, hybrid real-plus-AI footage, and traditional production with AI only in pre-production.
04Editors and ideas people
  • Editors who can also generate and manipulate footage now control the whole story, which makes them far more valuable than before.
  • Experienced traditional creatives get dramatically better AI results than beginners because they already know what makes a good image.
05Taste and where a young creative should start
  • Taste comes from lived experience, and bringing those specific memories to the prompt bar is what separates your output from everyone else's.
  • Learn one traditional skill first, then apply AI to that skill; aim for roughly 80% human input and 20% AI.
06Upskilling a team
  • Give your team 40 minutes a week of goal-directed tinkering with AI tools, and let whoever finds something useful teach the rest.
  • A dedicated creative technologist who tests new tools and filters hype can keep a whole team current.
07The copycat problem
  • Watching everything in your market is how you know what has already been done and what still feels fresh.
  • Copying with AI is trivially easy now, but copying tends to get you to second place at best.
08Remixing viral videos and letting fans design
  • Swapping a recognizable mascot into an already-viral clip borrows proven attention, but only works if your brand is instantly identifiable.
  • Letting fans design products with your IP, vote on them and earn royalties turns your audience into marketers.
09Jake Paul, the World Cup and brand risk
  • Licensing a likeness or inviting fan-made AI content can generate enormous free reach, but only if you can tolerate content you don't control.
  • Brands with a lot of reputation to lose should move carefully, while small brands with nothing to lose can experiment aggressively.
10The finished commercial and finding references
  • Your AI output is only as good as your references, so dig past Pinterest into deeper sources or photograph real places yourself.
  • Start a project from a single frame that grabs your attention; if it stops you scrolling, it will likely stop your viewer too.
11ShotDeck, Milanote and casting with Claude
  • Use ShotDeck's camera body, lens, film stock and palette data in prompts to keep continuity the way real productions do.
  • Cast AI characters deliberately with specific traits, because vague prompts return generic faces.
  • Use an LLM like Claude to turn your rough description into a detailed image prompt, then paste it into an image model.
12Handwriting the script
  • Write each script line like a director on set: shot type, camera position, duration, ambient sound and the emotional delivery of every line.
  • If you can articulate an emotion clearly, the model can usually generate it, so precise adjectives matter.
13Live character sheet of the host
  • Build a character sheet from many real photos (front, side, back, details) so the person stays consistent across camera angles.
  • Try the same prompt on several image models; in the demo, GPT image matched the host's face while Google's model did not.
14Product sheet for the shoe
  • Every object that appears on camera needs its own reference sheet, including inside details you may want to reveal.
  • Products with big, clear logos survive AI video well, while small text like ingredient labels tends to morph.
15Location reference and iterating
  • Treat the location reference as the color grade of the whole video and borrow camera language from a frame you admire.
  • When an AI image is close but wrong, update the original prompt and regenerate instead of editing, because each edit degrades quality.
16Voice reference
  • To match a real voice, record about eight seconds of a phonetically complete sentence in a flat tone and attach it as a reference.
17Writing and running the Seedance prompt
  • Fit only about three beats into a 15-second generation, or the model will lose track of what is happening.
  • Add a global look block to the video prompt so every shot appears to be filmed on the same camera and lens.
  • Tag each reference explicitly in the prompt so the model knows which image is the character and which clip carries the voice.
18Results, mining and recap
  • Generate multiple versions, check for hallucinated details, and stitch the best seconds of each together in a normal editor.
20Quickstrike questions
  • Budget real money: a 30-second top-tier generation is around $30, and full projects often need hundreds of generations.
Glossary

Terms worth knowing.

Generative AI (Gen AI)
AI models that create new images, video, audio or text from a prompt or reference material, rather than analyzing existing content.
Hybrid workflow
A production approach that mixes real filmed footage with AI-generated or AI-modified elements, such as changing props, makeup, lighting or locations in post.
Character sheet
A single reference image showing one person from many angles (front, side, back, top-down, close details) so an AI model can keep their appearance consistent across shots.
Product sheet
The product equivalent of a character sheet: multi-angle, detailed renders of an item so its logos, materials and shape stay consistent in generated video.
Aggregator
A platform that gives access to many different AI image and video models under one subscription, so a creator can compare models without separate accounts.
Morphing
When fine details such as small text or labels warp or change shape from frame to frame in AI-generated video.
Hallucination
Details an AI model invents that weren't in the references, such as a tattoo appearing on the wrong hand.
Mining
Running several generations of the same prompt and cutting the best seconds from each into one finished sequence in a conventional editor.
Creative technologist
A team role dedicated to testing new creative and AI tools, separating useful ones from hype, and teaching the rest of the team.
Whitelisting
An arrangement where a brand runs paid ads through a creator's account or licensed content, which on Meta platforms lets the brand boost creator posts as ads.
Global look
A block at the top of a video prompt that fixes camera body, lens, film stock and grade so every shot appears to come from the same camera.
Resources

Things they pointed at.

33:05toolShotDeck ↗
33:05toolCosmos ↗
37:14toolMilanote ↗
40:02toolClaude ↗
41:01toolOpenArt ↗
41:01toolChatGPT image model (GPT image)
41:01toolNano Banana (Google)
41:01toolSeedream
1:17:32toolSeedance 2.5
25:49toolSora 2
09:50toolMidjourney
43:47productWispr Flow ↗
29:54productAppLovin ↗
1:40:10productGenHQ ↗
1:36:40bookHarry Potter and the Goblet of Fire
1:38:20channelRoberto Nixon (RPN)
1:38:20channelColin and Samir
Quotables

Lines you could clip.

07:17
“Investing in a good video editor now is like investing in early Bitcoin.”
contrarian, one line, no setup→ TikTok hook↗ Tweet quote
07:20
“You can clip this in five years time, video editors will be one of the most expensive people on your payroll.”
bold prediction, self-aware 'clip this' framing→ IG reel cold open↗ Tweet quote
06:23
“The editor is the new creative director now.”
short provocative reframe→ newsletter pull-quote↗ Tweet quote
12:12
“You want to be using 80% human, 20% AI to get away with using AI.”
concrete rule of thumb with a number→ TikTok hook↗ Tweet quote
11:41
“AI is just the supercharger. It's not the replacement.”
tight, reassuring for creatives→ newsletter pull-quote↗ Tweet quote
1:03:51
“Your location locks in the color grade.”
non-obvious practical tip→ IG reel cold open↗ Tweet quote
1:09:08
“Every time you edit an image with AI, it effectively washes it.”
useful cheat code, stands alone→ TikTok hook↗ Tweet quote
1:27:24
“Kill one myth about creative AI. It's easy and it's cheap. It's not either of those things.”
myth-bust with dollar figures following→ TikTok hook↗ Tweet quote
1:34:41
“What percent of AI creators online actually know what they're doing? Three. 3%.”
shocking number, instant debate bait→ IG reel cold open↗ Tweet quote
47:45
“If you can articulate it, then you can generate it.”
quotable principle→ newsletter pull-quote↗ Tweet quote
Topic Map

Where the conversation goes.

01:13 – 06:52denseAI jobs, brand strategy and the three ways to use AI
06:52 – 14:11denseSkills that matter: editors, ideas, taste, traditional craft
14:11 – 17:23steadyTeam upskilling and staying current
17:23 – 30:15steadyCopying, remixing and fan co-creation
30:15 – 1:13:20denseLive build: references, script, sheets, voice
1:13:20 – 1:26:20denseLive build: video prompt, generation and mining
1:27:36 – 1:41:46steadyQuickstrike questions
The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogystory
Thank you for watching.
own proven system for avoiding slop and explains why he thinks AI will make more jobs for creatives, not less. And I'll tell you, you can clip this in five years time. One of the most expensive people on your payroll will be.
All right. Two things, guys. AI is coming and content fuels every single company in the world.
I am sitting across the table from a 27 year old man that has gone broke twice. is now doing eight figures a year and has one of the biggest AI creative platforms in the world. Stay to the end, and we're going to actually do a live demo on how he creates some of these videos, A to Z, from idea to end execution, something that you guys can take.
I want to start with three wild stats that you gave to me. One, the World Economic Forum says AI will create a net 78 million jobs by 2030. That's the headline, but underneath it, a ton of jobs are going to get destroyed, a ton more will get created.
And the people who win are the ones that are going to kind of reskill. Second, by 2027, 90 % of organizations expect AI -related skill shortages. And third, AI -fluent workers can earn up to 56 % higher wages.
So who actually wins with kind of all of these new AI tools? It's an interesting one. So we do live in like unprecedented times right now where this new technology has come in.
And so many companies are trying to figure out whether or not it's for them. And it sounds ridiculous because we all think that like every company should be adopting this technology right now. But there is a right way to do this and there is a wrong way to do this.
And most companies right now are doing it the wrong way. So I kind of want to set the record straight that most companies use AI in a creative sense for cutting costs and to turn over projects quicker. And I think a lot of people, you know, we know that AI does that.
We know that AI can achieve those things. But companies who focus on that end up pissing off their audiences so much because audiences, we're humans. We see the effort that goes into creative projects.
And that's what we love as creatives. So in my opinion, companies who win right now use it in a way that they know their audiences. they know where it's going to work and where it won't work.
So that's like the first question that businesses need to ask themselves. Yeah. I mean, the big thing that you told me off camera is like, you can use Gen AI and AI creative, however you want to use it.
It doesn't even have to be forward facing either. Like one of your big clients is just using it in pre -production, ideation, mood boards and such. Yes.
So we did a consulting session with the North Face. And when we went in there, it was interesting because They're a sustainable company.
They have athletes who are climbing to the top of mountains. It makes no sense for a company like the North Face to start generating AI footage or AI clips of their athletes at the top of mountain because it completely just takes away from what the brand is. So you need to understand as a brand, where can we leverage it to actually operationally save us costs like pre -production?
Is that creating the shot lists? Is that... building out everything that we need for pre -pro so that when we get on set, it's going to be done much quicker.
There's so many different ways like operationally that AI can help you in the creative process. I just think most companies are missing where it actually needs to be because the answer is not always in front of the public. It might not be in front of the customer.
Can you just walk through, you kind of spit off some of them, but I just want to double click and kind of... give everybody out there especially operators ideas on how they can use this and ultimately like what is gen ai generative ai and do you have to use it fully can it be hybrid just like give us a little backstory on it yeah so there are different ways that you can use generative ai So one example of a project might be a full gen AI project, which is from start to finish, you're writing prompts, you're creating something fresh out of an AI image model or fresh out of an AI video model.
And that's the only footage that you use in the project. That's a full AI project. And then you've got on the other side of the spectrum, you've got things like hybrids, right?
And hybrid workflows is basically when you're using real footage mixed with generative AI. So that's still using human craft. That's still filming with cameras.
That's still doing location scouting. That's still doing a lot of the things that you would do in traditional, but you're using AI to augment and improve upon the original footage. So that might be VFX, for example.
And then you've got on kind of like the full other end of the side, you've got still traditional, but then you're just using AI for pre -production. You're still using full real footage, still full real actors and all of those things. no synthetic actors as we call it, and you're just using it for pre -production.
At the absolute minimum too, we're going to pop up right now that video of you like swinging a baseball bat and then you're bringing him into this world and you kind of, I don't know, you like digitize him and then you give him the ability to kind of do additional things and add on more animations. Yes. So the example that you're seeing right now is We filmed this in New Hampshire in like a completely remote spot that was like an abandoned shopping center.
And we bought props from Dick's Sporting Goods and we basically used those props to reimagine what it could look like if we wanted to film this. as a Walking Dead style shot, right? So we use the slow -mo feature on the iPhone to get the slow -mo shot right at the start.
And everything was just using my best friend. And we literally just like created this scene, took about 50 takes to get the right one. But it's using AI with real footage to adjust props, adjust makeup.
applying all of the zombie effects on characters, changing the location slightly. There is a lot that we can do now. Lighting, we can change all of this in post -production, which is why a lot of people say that the editor is the new creative director now because editors who have the ability to edit footage together can now generate the footage as well.
And they are the true storytellers in a lot of traditional work because they're the ones who piece it together to make it feel like the full project. And they've got full control now. That's a pretty hot take that editors and new creative directors.
I want to dive into that. I mean, one of the things that I was going to ask, and I guess it kind of falls in that bucket, is what skills are more valuable and less valuable in this new age directly related to just creative? Yeah, so I'm going to give another hot take here.
So I've been saying this for about a year, but video editors investing in a good video editor now. and this means a lot of different things to a lot of different people, is like investing in early Bitcoin. And I'll tell you, you can clip this in five years time, video editors will be one of the most expensive people on your payroll.
They are going to be so valuable. In my space, like right now, it is very, very difficult for me to hire a video editor under six figures. Like it is very, very competitive.
And what a good video editor is, is also changing because video editors now, it's not just using tools like Premiere Pro and After Effects to create really cool projects. It's actually a video editor now is someone who is hybrid. They can use AI.
They can generate footage. They can manipulate footage. They can piece it together in the edit.
they have full control. So photographers, videographers, editors, directors, I mean, there's just so many different things. Outside of editors, who do you think is best positioned in the kind of the creative space to ride this wave moving forward?
Is it the ideas person, specifically that? It leans into the ADHD ideas, people, because we spoke about the ideas being the currency of the future. And I want to go into that a little deeper because When you have this abundance of content that we spoke about and everyone can create what is deemed quote unquote beautiful imagery or beautiful video, but it doesn't mean anything.
You know, you can create something, you know, a single clip or a couple of clips stitched together. But it doesn't mean anything. To put like a full creative project together takes a story.
It takes narrative. It takes a lot of ideas. And a lot of people who are in AI right now are people who love getting quick wins.
And so the traditional creatives that we've worked with, when we've gone and done workshops with people who have worked in marketing for years or people who have plenty of experience in their field already, and we give them access to tools like Midjourney to create crazy imagery, the stuff that they come back with is 10 times better than what I can make because they have the photography background.
They've got all of this knowledge of what constitutes to. good image they understand taste and obviously taste is a word that's thrown around in our industry a lot you know what is the key differentiator in taste and i'd love to know like in your opinion like how does taste make an impact in in the world of ai now yeah i mean i've asked a lot of people across this table and i keep getting different answers and it keeps iterating i ultimately think taste can't be taught and it goes back to a great episode we just put out with ronnie feig from kith i think he explains it best of just like your formative years, I would say like 12 to 18, the sports, the culture, the music, and that is embedded in you forever.
Like what you loved as a kid, you're going to love forever. So I don't think taste can be taught. I completely agree that the ability to storytell and having not only good taste, but also consistent taste is another thing that I've learned is staying narrow and disciplined.
You like what you like. You have to believe in what you like and then stay with that. Everyone has their own.
unique taste. And like, it's so funny, man. When I, when I look at you, when we talk, when I meet you, when I look at your videos, like the video that you sent me will pop up right now of you and the homie getting the suntan lotion from like the old woman.
Like that makes sense. Like when I look at your taste, like funny is a part of your DNA. So I think it's just being true to yourself.
Yeah. And I want to talk about this because in the world of AI, taste is truly everything. And It is your human experiences as you said is what develops your taste.
It's the things that you've experienced in your life. And it's that human key differentiator. When you arrive at the prompt bar, it is that holiday that you went on with your family, maybe when you were eight years old.
It is what the stones felt like on the beach when you went to the beach for the first time. It's these subtle things that have enriched your life that when you arrive at the prompt bar, you bring those ideas to life. And it's being able to articulate well, because...
Everything in AI is about prompting. It's about being able to get the idea that you have in your head out into a little black box, which will then give you a nice image. And a lot of people, when they start out in AI, they don't understand that concept.
So it's about being able to articulate what you want. For both the operators to know from a hiring perspective and from an upskilling internal team perspective and then actually creators, if you're a 25 -year -old creative, where should you start? from a skills perspective in Gen AI?
Like, what should you know first? That's a great question. And I actually think the answer is not what people would expect.
I think the answer is that you need to have a solid foundation in the traditional first. So if you're 25, you know, you're young in your career, you need to understand something like photography or video editing. You need a traditional skill.
AI is just the supercharger. It's not the replacement. It's the thing that you use to...
like literally take you to that next level. And when we tell people about like creatives about using AI, the way that we say it is that you want to be using 80 % human, 20 % AI to get away with using AI, right? So let me break that down for you.
When AI first came out, right, you could get away. with a 50 % human input, 50 % AI input. What does that mean?
That means you could use AI maybe in the early days to write a script because no one had seen AI written text before and it was new and it was novel. Now, everyone's seen ChatGPT written emails, everyone's sick of them. Like it's, you know, text is rubbish.
It sounds awful. But as we as a society get more used to AI, we get more used to seeing slop. We get more used to seeing crap.
And so what we... have to do is the amount of AI that we use in the creative process is getting thinner and thinner, in my opinion, to get to a project that people actually resonate with. So right now at our company, we are saying like 80 % human input in terms of ideas, execution in the edit and things like that.
We're using AI to do like 20 % of what we're actually spending our time on. The generation side of things, you know, image and video gen takes 20 % of our time. The 80 % is still the human side of it.
But to answer your question, creatives now, if they were getting into AI, think about what you're good at. So for me, in my early days, it was Photoshop. and editing pictures.
That was what I was passionate about. So I started to look into how I could use AI to relight my images. That was one of the first things I did.
So take your creative profession that you do now and then start seeing how you can use AI to, you know, if you're a camera operator, you know, what can you use AI to enhance what you're doing? So it's about identifying what that is because AI impacts every part of the creative process now. There's something that you can use.
that will change the way that you work. Makes sense. I want to tie a bow on this and just go and flip to, we talked about the creator side.
Let's just talk to the business side of it. For an operator out there that wants to kind of upskill their team outside of joining your platform and kind of learning, but let's double down on that and really understand what is the first thing they should do to upskill in relation to AI. Yeah, first of all, I think upskilling your team in AI is probably one of the best decisions that you can make.
Immediately would make them more valuable, right? So what you can do now, like a tangible thing that you can carve out is 40 minutes. And this is the 40 minute rule.
I did an interview with the head of Google Flow, who is responsible for releasing all of Google's top AI tools. And I said to him, internally. how are you and your team keeping up with AI?
In a world where everything moves so quick and all of us outside can't keep up with it, how are you and your team doing it inside? And he said to me, we carve out 40 minutes every week and they call it tinkering time, right? Tinker, play time.
Release your inner child and let your creatives in your team play with AI tools for 40 minutes, but play with intention, okay? So play with a goal. in mind.
So we're going to create a commercial with AI by the end of the month, and you're going to have 40 minutes a week to play with some of the tools and see what you can put together. Then if somebody in the team does something that you're impressed with, they can put their hand up and go, I'd actually like to run a session on this, and I'd like to teach people what I've learned, and then diffuse it like osmosis inside your team.
So what we have at our company is a creative technologist. And this person's only role is to tinker 24 -7. They are full playtime, play, play, play, learn what is coming out with all of the new AI tools and then teach everyone internally.
And they run a session every week and they keep us up to date. Their job is to go and look at the market and decipher between what's bullshit and what's actually worth using. And then they only teach us what we need to know.
And having that role is massive. I was just going to say, that seems like a very valuable job, the ability to pick up any tool and decipher what actually is valuable and what isn't. Huge.
And I spent time with Mr. Beast recently, and it was one of the first highlights that he wanted to make after we had chats. It was like, you know, he needs someone to come on set and be like, okay, we do a crap load of VFX.
You know, maybe you are in the creative space and you do a lot of work. Having someone on set. as well, who can tell you this is probably possible with AI.
You know, we can film this, but we could do the effects with AI to kind of like save us a bit of time and money. Having that person on set saves you a lot of time too. So it's like having someone in the background who is learning and teaching internally.
whether that's operations, systems, creative, whatever it may be, but then also on set as well. It's very helpful to have someone who's got AI knowledge. And I think when the stakes are as high as something like Beast Games, having that, you know, maybe even fucking three people on set to be like, hey, we probably shouldn't do this.
Let's save the time and do something else. It's interesting. I also want to talk about AI and YouTube because as somebody who is very, I can spot AI from a mile away, right?
I would probably say that I'm I have consumed more AI content than I would say nearly anyone on the planet. And this is really interesting to me.
I spend so much time watching and consuming because that's how I know I'm making good stuff. Because I know. the database of videos that exist in my world in the same way that you probably watch a crap load of podcasts and thought, how are we going to differentiate ourselves?
How are we going to make us, you know, our content look different to our competitors and all these things. You need to know your market. So like somebody who is a creative, a good creative now needs to understand.
all of the data that exists in that market and what their competitors are doing, because we always want to elevate. We want to be the best that we don't want second place. So, you know, do you consume a lot of other podcasts?
I'm more just seeing who's copying and that just gives us the signal to iterate onto the next thing. Yes. Let's talk about copying.
Yeah. Because AI is probably the most powerful tool for copying. And I fucking hate that.
Okay. Do you know what's so funny is I make educational content on AI. The amount of fucking copycats there are that quite literally will take my exact video and just put their face over my face and just make my fucking video is unbelievable.
And I can't get mad about it because I'm teaching people how to do this shit, you know? So let's talk about copying because now more than ever, it is easy to copy work. And like we can throw some videos on screen and I actually sent you some examples of videos where people will create a video and then you will see that they have replaced the character in that video with a different character.
Right. And I think that there is good copying and bad copying with AI right now. And I think there is a huge opportunity in the good copying section.
So let's dive into like good and bad, if you don't mind. I would love to hear it because for me. I'm not emotional to it as well.
Oh, I get emotional. I get pissed off when I see someone copying my shit. This is my second go around with my art company.
I had a media buyer take our customer list, our email list, our files, our Facebook account. Yeah, this was late 20s and that was a different mark. For me, I'm more just like, bro, like just change the color, like change it a little more.
Something. And again, kind of going about what we talked about off camera before is just like. I don't know, man.
I just think that no matter how good this gets, I still think we're far away from being indistinguishably like the same, like you'll always know if it's, if it's copied. For sure. Like, I'll tell you a story.
I was in Dubai at a conference and a creator came up to me and he said, Rourke, I just want to say thank you. And I said, bro, for what? Like, what have I done?
And he goes, like, he's like, I. I watched your content and I started making AI educational content because of you. And I'm here with Google Pixel and like they paid for my flight here, sponsored me to be here and all this stuff.
And I was like, dude, dapped him up, you know, gave him a hug. I was like, man, that's awesome. So happy to hear that.
That's like your actual mission in life. Yes, literally. My mission is to create 100 ,000 jobs for creatives.
Yeah. So I was like, I'm doing the shit that I'm meant to do. Right.
And then he's like, dude, check me out on Instagram. And I was like, cool, cool, cool. I know where this is going.
Yeah. And I and I checked him out in the hotel room later. I went this motherfucker, like the literally my visuals, my face on the videos.
But he had just changed the talking head of him talking at the bottom to him, my scripts, my music, my everything. And he had managed to get there with Google Pixel. You know, and people are building whole careers off of copying.
And that's not that's been the case for a long time. People have copied their way to the top. A lot of people have.
I need to correct you, though. In other words, Caleb Ralston, you're going to copy to second place at best. Yeah, I'm going to leave it at that.
Yes. OK, maybe not to the top, but like people definitely copy their way like up, you know, and I don't think it's like necessarily fake it till you make it. But it is it is a problem now.
And we should discuss it because this is one of the negative sides of Gen AI that I don't like. And there's not really a lot that we can do about it now because it just is so easy. But I think there is an untapped opportunity that I think many businesses can tap into here.
And we should throw up some visuals on screen. This clip went viral on Instagram recently, and it's this old guy who's basically saying, you know, you're not allowed to fish here, like this is state property, like you can't fish here. And, you know, it's a comical video, like this guy's got the greatest accent of all time.
And what was really funny, that video got 2 million likes off the bat, right, from the original creator. And then what we're seeing is people changing that character. with AI to be someone else.
And that video will get half a million likes, 200 ,000 likes. And then someone will change it to be a celebrity, like a homelander from the boys. And then they'll change it to be all these like Hasbulla and like all of these like funny internet characters and personalities.
And what it's doing is it's creating more viral content from the original content. And I think that there is a huge opportunity for businesses to tap into this. So let me give you an example.
Duolingo, a company like Duolingo, replaces the old guy with their mascot, right? Now, Duolingo posts that video, they're tapping into internet culture. They're tapping into, let's say, meme culture or social culture for brand relevance.
And it doesn't work for every brand, but for those brands who... are trying to tap into internet culture, it's a very, very powerful tool that you can use to associate your brand. But in order to associate your brand, you need to have strong brand awareness.
You need to have something that people associate, you know, with your brand. So something like Duolingo where they've got like a popular mascot is pretty easy to associate with the brand. So this goes back to POV and ideas.
Yeah. Like you have to come up with a big idea, like someone that thought Hasbulla would be funny in that specific scene. Yeah.
That's basically taste, although that's kind of like an easy one to do. But, you know, picking what makes sense in that setting. And it's interesting because the video has already gone viral before.
So it's guaranteed to perform, kind of. It's like the freest marketing that you could do. Now, use AI as a tool to build fandom.
And this, I think, will be a huge, huge thing in the future. So a couple of things. I did a consulting session with fans.
right Vans Vans yeah and um accent bro I thought you said fans so many so many times on this he said brown before I'm like yo did he just say I was gonna I was gonna ask you the fucking accent accent please baby yeah so Vans I did a session with them and when we were talking through like how Vans could use AI, right? It was like, okay, well, companies like Vans have allowed their audiences to create custom shoes before, you know, with their shoe shape, but you could design your own shoe and then have it made and then wear it.
Custom, feels cool, like feels like you've had a play in the product. Great. So I said, why not let your audience, give your audience the flexibility to use your IP?
And use the Vans brand to use generative AI to create designs for t -shirts, for shoes. And then on your website, you could have a section where people could upvote great designs and things that they actually want to get made as a real product. So you have like social approval in there.
And then the best products that people actually go and make. that person who designed that product gets royalties in the product. So now you have your audience who love your brand making things that they want to see for themselves.
And if it goes into production, they have royalties in it. And imagine that headline of someone saying, first person to make a million dollars by using AI to like... design something like a t -shirt.
That's crazy. That's marketing right there as well. Like, I think that there is so much opportunity to let your audience become a marketer for your brand with AI.
And there's a couple of interesting case studies that we should talk about here who are already doing it now. Jake Paul is one of them, right? Like him, love him or hate him, right?
When Sora 2 came out, Sora 2 is a video model that was released by OpenAI, the company who made ChatGPT. When Sora 2 came out, Jake Paul was one of the early people who, I think he's an investor, but he was one of the early people who licensed his likeness to Sora.
So anyone could create videos with his likeness. And he got... billions of views with his face on it in the first, I think it was like the first week of Sora 2 coming out.
That is free promotion for his brand. We'll pop that up. He literally took over Instagram.
He legit, like it was him putting on like people, like him putting on makeup and it was like him doing shit that he would never do. And like this surprise factor, some people might look at that and go, you know, that's probably a bit embarrassing. I don't really want that look for me and my brand.
But for people who don't care and realize that attention is currency, they win. Like Jake Paul won in that situation. He got free marketing for that and billions of impressions and views, which all go back to, you know, his deodorant or whatever it is that he sells.
So it's like, it's an untapped potential for businesses to take advantage of. And we'll talk about the World Cup, okay? So this year, we spoke about this before we recorded, but I think it was 5 % or 6 % of clips that were uploaded on social media this year about the World Cup were AI generated.
And out of that, it was 5 billion views, I believe it was, of AI -generated clips from the World Cup. And now think about what that does for fandom, right, for people. People were making videos of Erling Haaland looking like a Viking.
People were making videos of Kylian Mbappe, you know, jumping off the statue in Rio de Janeiro. You know, some videos getting like 50 million views a pop. And all it's doing is creating fandom for the World Cup, where you're letting your audiences create in a way that that was never possible in the past.
And like, you know, as attention is the currency, there is a massive opportunity to leverage AI as a way to create fandom right now. Do you see a world on the flip side where this stuff kind of kills the authenticity or could hurt the brand? Massively.
And that's why it's dangerous. And I am a... I am an AI for good person.
I truly believe that like, you know, there are good apples and bad apples in every industry. And I think that like with AI, it's an incredibly powerful tool. And with that comes a lot of responsibility.
So it depends how much you also have to lose, right? If you are a company who has a lot to lose by damaging your brand reputation, then it's probably not the play, right? Whereas if you are a small startup who, doesn't necessarily have an existing, you know, name to yourself or a brand, you've got nothing to lose, baby.
Fucking go for it. Like there's an opportunity there to just like let rip and build something crazy with AI. But yeah, I think like for a lot of brands, as we said at the start of the pod, there is that spectrum where you need to understand where do we want to use full AI, partial AI, or like, you know, where does it fit for our company?
Good to know there. Wayfair, Hexclad, and Ridge are already scaling with it. Head to applovin .com slash open residency.
If you spend $5 ,000, you get $5 ,000 free in ad credits. That's applovin .com slash open residency. All right, guys, we're going to go through A to Z, how to create a hyper -realistic brand video using generative AI.
So first, we're just going to pop up the video for you guys to watch first to just kind of set the stage. And then we're going to start from idea all the way through to completion. So let's get into it.
Cool. My AI agents can do your entire job. I don't even need to work anymore.
What's the fucking point? Of what? AI agents.
Like, why the fuck are we even building them? They do your job for you, bro. So you've got more time.
More time for what?
Take that Tamagotchi, for example. That thing gives you a purpose. That's why you love it.
It needs you. Okay. So imagine I gave you an agent that played it for you.
The agent feeds it, cleans it, keeps it alive 24 -7. You'd never touch it again. Look, creatives don't want an agent to make something for them.
The whole point is making people feel emotional about our work. And now we're handing that to a machine that's never felt a thing. Good ideas come from the mind, not a machine.
Exactly. AI can be used to help tell your story, but not automate the whole thing. The thing comes from within.
Anyway, I don't want more time. I just want something to do with it. Cool.
What do you think on that one? A masterpiece. I've seen a lot of your work.
I love that. That hits the heart. Yes.
Let's start from you wake up. You want to create this video. Yep.
Where do you start? So first thing that you want to do is you need to find references. So when you're making an AI project, your output is only going to be as good as the quality of references that you find.
You need to know where to look to find high quality references. Platforms exist like Pinterest, Cosmos, Shopdeck. These are a couple of different options that you have.
Can you basically get inspiration or references from anywhere? But you're just saying like the search functions and the volume and quality. Yes, it's definitely like volume and quality.
I think like, okay, platforms like Pinterest, everyone goes to. It's kind of like a bit cliche. So I actually avoid Pinterest entirely because when I look at what other people in the AI space are doing, I'll like see something from Pinterest that I know somebody's nicked and I'm like, hmm.
It just isn't right. Like I like to go to the deep depths of the internet. And like when you are true in your craft, the reason why people have taste, in my opinion, is because they've spent hours in that specific field.
If you love motion blur, you're spending hours looking at different, you know, videos, images that have motion blur in it. And you get so deep in the references, you understand what constitutes as good. Hypothetically, though, if you're in a little street in Tokyo and in a vintage magazine shop, you can take a high resolution picture of that and use it as a reference point.
You absolutely could. Yeah. So I've done projects where I've literally gone outside my front door and I've taken reference images of the little street that's outside my house and I've made a project and we can throw it up on screen right now.
It's literally the whole street. outside of my house. So if you have a specific location that you want to go and scout out, you can take a bunch of reference photos of that location and then use those as your references to build out your scene.
Love that. So you have, I guess, I guess, what about idea? So you're having the reference points, but is there an idea first?
Yeah. So I think like in the early days, a lot of creatives struggle on the ideation side. And I think like platforms like ShotDeck are a way that you can help you come up with ideas and actually Cosmos as well, which is another, it's like Pinterest for designers, basically, where you have a lot more rich photography and imagery for inspiration.
So ShotDeck, this platform here. is basically frames from movies. And you can scroll through here and it's got a bunch of different references that are really, really nice frames.
And what's interesting about Shop Deck is I treat it as a social feed. When you're looking through all of this, like what stands out to you? Like I'll ask you now, like what frame when I'm scrolling through here, is there anything that like jumps out to you as like, oh, that's called my attention?
Yeah, like if you go up a little bit. Yep. Like right there, it looks like a Jake Gyllenhaal where it's in like a serial killer vibe.
Right. Yeah. So shots like that, I immediately, if that's taken your interest, then you know that there is a load of images that we're looking at here that are all competing for your interest.
and that one stuck out to you. So if it stuck out to you, it's probably gonna stick out to someone else who's watching the content on the other side as well. So this is a really nice way that I like to, when I'm starting a project, I wanna start with a frame that's gonna capture your interest.
So when you are looking through this, I'm treating this as my competitors. All these other images are fighting for my attention, but which one's calling me? So I will then like click onto any of the images.
So let's just say it's this one right here. Right. We've got this nice close up shot.
What's fascinating about shot deck is it breaks down all the colors of the frame from the color grading into almost like, you know, a color palette here. And what's also super interesting is all of this data that you get about this shot. So, you know, the director, the cinematographer, you know, what film stock was used, the lenses, the camera body that was used.
And all of this data is so key when it comes to AI, because when A lot of people, when they make AI projects, they forget that... you know, in traditional, when you're filming a commercial or you're filming a project, you're using like similar camera gear from shot to shot.
And that's what provides continuity between projects. And people seem to forget that when it comes to AI as well. Because if you are specific about the style of camera, the camera body, the lens, the coloring, you know, all of these core details, the composition, all of these things are critical to the output of the image that you create.
So now, with Shotdeck You can use it to help you like build compositions and we'll come back to this a little later.
But you can use this and just scroll through, find really cool references. And when you find one you like, if you click on here, you can actually scroll down this platform and it will show you similar images as well. So there's tons and tons.
It's like an endless loop of just inspiration. So this is a great place for you to just get ideas because you might not have an idea for a video or a commercial. But if you find a frame that you like to begin with, with, that might now dictate how the rest of the story goes.
So for example, here, this is such a cool frame. It poses so many questions to the viewer. What am I looking at here?
What is the context of this scene? So finding something that catches your attention is a great way to just like help you come up with concepts that you can then go and like dive into to start building on top. so this is a good starting block it's a great like just think tank in general like i obviously like this is very very nuanced and you know you do a lot of kind of short form stuff but i just think from like an ideation perspective if you're like an e -com brand or some sort of consumer company like you can get ideas actually just through going through this yeah totally so Collecting references is the first step.
And what I like to do is I like to just go through and download a bunch of different frames that just like capture my attention. And usually I open them and I'll put all of them into a platform like Milanote. Milanote is a platform.
So if I go into my documents in here and I just like drag and drop a bunch of photos in. you can see that this is basically like a mood boarding platform where you can like add your frames in here. You can add in like notes on top of things that you're interested in.
It's a platform where you can basically like you can annotate, you can like make notes. And we use this all the time. So just to give you a bit of context, like these are like a bunch of different projects that we've done.
I can double click into kind of any of these. And this is like where we're just like visually dumping a bunch of things that we're like interested in. And is this you're like clustering?
different scenes. It's ugly. It's the ugly just getting things together so that we can start organizing some kind of concept.
And you can get really organized with this kind of stuff. So a good example of this is right here. So when you're delivering AI commercial spec work, you want to make sure that you're being very granular.
So what we'll do in here is we'll actually like almost do like a character casting sort of thing. Because when you look at each of these characters, they give you a different vibe, a different emotion.
And a lot of people fail to think of casting characters in their projects because AI gives you the most generic looking BS, to be honest, when you don't get specific. But if you try to like target specific nuances about characters here, like I love how lost this character looks, like the bed hair, you know, it tells you a lot about who she is as a person.
Just for context too, is that a ref file that you got from the past website? and then now you've optimized it through prompts or is that? No.
So so this will be just like me going into a platform like Klaus, for example. So inside here, what I'll do is I'll say like on a white background, can you create like a reference image of a character who's got like a red turtleneck sweater on? She has like slight bed hair, but she looks kind of composed, but also like she's kind of late for things a lot of the time.
She looks like the kind of lady who'd be like a mother of two and she. She's a little bit lost in life, but she hasn't got a partner and she embodies this like floating character who's just like floating by through life. So like giving these kind of context of things helps me like picture these characters in my head.
And you can say like, write me a prompt that I can use to generate an AI image that I can copy and just paste into an AI image generator. And now. Claude will take the context of what we've done and it will start writing a prompt, which we can then go and give to an AI image model.
Under the notion that Claude doesn't have the best AI image generating model, you'd use other ones. So Claude doesn't actually allow you to generate images specifically with it, but we use Claude as our prompting process. So you use Claude like your partner to help you articulate what it is that you're trying to make.
And then that's going to be the raw input to then go into the image generator? Correct. So this like mega paragraph, I don't have the time to sit and write all this detail, you know, but it's gone in and it's written all of this like amazing rich context in here.
So I can just go and copy that and we can go to platforms like OpenArt where you can just paste it into the AI image section and we can generate an image. And we can generate multiple of these images. So we could do like five at a time and we can just start generating.
When it comes to AI as well, you have different AI models. So in here, you can click in here and there are a lot of different AI models that have come out over time. And the main ones that you really need to know about is OpenAI's image model.
So ChatGPT's image model, which is ChatGPT image 2 .5. One of the better ones out there, to be honest. You've also got NanoBanana.
This is from Google. Very, very reputable model. And Seedream.
These are like kind of the top three image models. And OpenArt is a platform that crowdsources and you can tap into any of the models? Correct.
So OpenArt is a platform. It's what we call an aggregator. And this is a very key term in the AI space.
So it's an aggregator that brings all of the top models together under one subscription. So, for example, I've used GPT here and we can see that it's already like created like this multi perspective of this lady in here where we've got like two different kind of shots of her. This is a starting block.
But what you want to do when you're making AI work is we want to create something called a character sheet or a location reference sheet or a product sheet. And I'll show you what that kind of looks like. So here is an example of what I would call a character sheet.
So this is me and we used a reference of my face to create me wearing these clothes. And to do this, like it's just as simple as uploading a reference image of your face and then also uploading a reference image of the templated. character sheet.
And what we'll do is we'll leave in the description a character sheet of my face that you can use as a template. So the way it works is... And just for context, this, what you just showed us, that is the input that's used for the commercial that we showed before this that we're walking through.
Correct. So to create the commercial and the video, super realistic video, you have to give AI ingredients. And we're going to give AI and what we're going to build is ingredients for characters.
locations and props. So we're really designing everything before we actually start doing the video generation. Video generation comes right at the end.
A lot of the AI work comes in ideation, in the generation of your characters, the locations, the props, all of that kind of comes first. So let's just go through this chronologically to recap. So first of all, you're going to find your references.
Second of all, we're going to go and we're going to write a script. because we need a solid idea right and then we can start creating our characters and locations and everything so scripts this is the least sexy part of the video and it's just in you know blank documents we need to write out the idea for the project And this is really, really key.
So the way that I write scripts for AI projects is different to how maybe movie directors or traditional would go about writing scripts. Mine is like, you know, and we'll dive into it right here. So let's take a look.
And is this done via Whisperflow here or is this done in Cloud with an empty file helping structure it? This is actually me handwriting this puppy. Like I am going in and I'm...
Oh, wow, guys. Yeah, we're going old school, all right? So yeah, this is me like...
closing my eyes and picturing the scene. I'm picturing exactly what's happening. So for example, for the video that we just watched, my script started with, it starts with a wide shot.
So I'm telling the AI model, what kind of shot is it that we're opening with? Is it an establishing shot? Is it a wide shot?
Is it a close -up shot? Is it a macro shot? What style of shot is it?
And that's, you know, your knowledge that comes from traditional of cinematography, photography. That's where all that comes in. I was just going to say, like, reference back to, like, the beginning of the podcast.
That's everything. Like, I don't know those things that you just said. Right, and it helps.
You learn it over time, too. I don't come from film, but you learn it through, by just playing with AI a lot, you learn that you need to have these things. And, yeah, it's just a process of learning.
So, for example, here, it starts with a wide shot of me and my friend sitting on the bench, and we can see it from an aerial view looking down. So I've now positioned my camera.
I'm telling the AI, where is the camera positioned? I'm also saying, what style of shot is it? And also, what are the subjects?
What's actually happening? You know, what can we see? And then on top of that, I've got the camera holds on the scene for two seconds.
So I'm now like picturing it in my head. How long is this scene playing for before it then cuts to the next scene? So I've got the camera holds on the scene for two seconds and we can hear park ambience, subtle bird sounds.
So we're describing the audio cues that you can hear in the scene as well. So really like place yourself in the environment. What can you see?
How do you picture the camera being? And what can you hear? And start just like putting it all down.
And everything will come together towards the end. But just like putting everything down on paper is the important starting point. Then you just want to like direct as though you're literally on set.
So for example, you know, for two seconds, it holds here. Then the camera cuts to a shot of me watching on my phone. And I'm reacting to Instagram reel about AI agents and how they can do my whole job for me.
And there's no need for me to work anymore. So I've just described. the exact scene that you can see like right here.
Then I paragraph it out and I just go bit by bit by bit and I brick by brick build on top of the story. Saying things like small cues, like I get visibly upset and frustrated. Being able to articulate emotion and having that like emotional understanding of how you and I.
act in certain situations, I think is a huge skill in AI. Like having that emotional awareness that you can describe how people feel or should be feeling and how that is portrayed is a great skill to have. Because if you can articulate it, then you can generate it.
Is the rule here, because again, like everything kind of ebbs and flows, is it just be as prescriptive as possible in everything, in the camera angle, the character, facials? The more context that you can give it, the better, right? So things like this, you know, like I love this, like make a frustrated noise as I throw my phone against the grass.
You hear a thud of my orange iPhone 17 Pro Max hitting the floor. And I say in a frustrated tone with a frustrated face, you know, describing how is the vocals being delivered? Not just saying he says.
You know, you're saying, how is that, you know, that line being delivered? These are all the 1 % that I'm giving away to the audience right now that will transform your projects from looking like crap AI to getting real human, what feels like human performance. It seems as though adjectives are very key here, huh?
So key, dude. So key. That's basically the process.
And once you've got, you know, your whole script written out, this is what you do to start creating like, actually, you know, we'll come back to that. a little later. So really quick, just to stop you here, a big thing that I'm going to want to ask.
And I just, just because I don't want to forget about it is just when we press the button to kind of generate it, like, are we going back? And like, if we don't like. The frustrated face in six seconds in.
What do we need to do? Like, do we need to redo the script? Yeah.
So we'll come on to that for the video part of the sequence. There's a lot of iterating and changing that goes on. So we'll come to that because that's a really key part.
So now we're in a position where we've got our scripts locked in. We've got our references. And what we're going to do now is go to the image generation phase.
So for every project that you do, you have to create character sheets. Now, character sheets look like this. And again, I will put like a link in the description so you can go and download like a reference of what our character sheets look like so you can use it as a template.
And is the rule of thumb here the same as before? The more, you know, descriptive, the better? Yeah, so let's kind of like walk.
Well, actually, yes and no. So what we can do is we could create a character sheet of you right now if you wanted. Yeah, let's do it.
Yeah? Yeah. Okay, I need to grab my phone then.
Let's just take a picture of you. I'll airdrop it to my thing and then we'll create a character sheet so that the people at home can also go and do it. Oh, boy.
Yeah. Here's the end? Yeah.
Okay. So we'll take a couple like wide ones and flick side.
Okay, so what we've done is we've just taken a bunch of reference images of Mark here. So we can go and upload all of these into Claude. And we've got macro shots in here, close -up shots, detailed shots of...
you know, tattoos and specifics in here, you know, back of the head, side profile. We've got everything. All right.
And so now what we need to do is we want to upload a basically a template of what a good character sheet looks like. And this is like an example of mine. And what I like to do is create a character sheet that has like front, side, back, you know, headshot.
top down shot, and also some detail shots of eyes, beard, all these kinds of things. And when we give AI all of that context, it knows, you know, when it cuts to a scene of you from behind, it's going to look the same. So we're going to get that continuity.
So what we're going to do is just inside Claude here, and you could use chat GPT, it doesn't necessarily have to be Claude. We're going to say, create a character sheet of this guy here. And I would like you to do it in the same style as the character sheet reference that I have uploaded.
Something as simple as that. Maybe we could say as well, we could even be more detailed and say, include detailed shots of the tattoo on the neck inside the character sheet. Write a prompt so that I can create this inside of GPT image too.
So what I did at the end there was I just gave it the instruction that we're going to be using the image model GPT -2, which is ChatGPT's image model, which is available on OpenArt. But I could equally say, you know, we're going to be using NanoBanana or Cdream, or we could, you know, say any of these different AI image models.
And is it good to say to like 6 '1", 180 there? Oh, dude, totally. Great shout.
So how tall are you? Six foot one, 180. He's added a few inches on there.
Six foot half an inch, 180. Six foot half an inch, 180. Okay, cool.
So six one, yeah? So let's just say like he is six foot one. So maybe add like a little scale on the side that shows his full height to be six foot one.
Cool. Alrighty. So now...
Claude is going to take all of those references and, you know, this prompt that we've given it, and it's going to go and draft a prompt for us that we can then copy and give to OpenArt so we can start building our character references for the project. So here we can see it's going through and doing all of that. It's got the tattoos in there, the jewelry, two -layered necklaces, gold rope chain, silver diamond tennis chain.
Wow, I picked up both of the chains. Look at that, dude. We've got all of this context in here.
Pretty crazy, including like all the details of... And you're saying that's enough without the reference images to put in? So this here, if we just use this, so what we're going to do inside of OpenArt, if I just pasted this as a prompt in here, it probably is going to create someone who looks relatively like you, but it won't be you.
And we should probably test it just so that you can kind of see because it will be good to see. Let's see one without the additional photos. Yes.
Cool. So while those are generating, what we can also do is provide references of what you look like now. So it actually has context.
So what we're going to do is we're going to click on in here and we're going to upload images. So we can click on image, upload images. I'll just take all those.
We can just take all of them, dude. We can just give all of these in here as references. And these models can take a lot of image like references.
So you can give a lot of data as context so that it has everything you need. And the same thing applies with products. So when you're making AI projects, you need to make character sheets so we know what you look like.
But then we also need references of the product as well. So all of these references are just loading in here. So we can go in and just click on each of these images and load those bad boys into the prompt.
Cool. So now we've got all of that in there. I'm going to make a couple of these.
Wow, look at this guy. Yeah, look at this guy. Wow.
Look at this Chad. That's a good hairline right there. Yeah, he's gone to Turkey for sure.
Yeah, that's a Turkish one. And what we can do as well is we can create like any aspect ratio of an image we want as well. You know, I typically for our character sheets like to do 16 by 9, just because you get like the wide context of everything you need.
You can adjust like quality of the image as well. And by adjusting quality, it adjusts like the cost of the credits. And I think cost of credits is important to talk about here because there is a misconception that AI is cheap when...
It is cheaper than traditional, but it's still cost associated with a lot of this stuff. So we're just going to run like a bunch of generations and hopefully we get something cool that comes in. Let's see what we got.
What I also like to do, which I haven't done yet, is add a reference of the templated character sheet from before. So for example, we'll go into my upload or images. And we'll upload a reference of what like a good character sheet looks like, which is this one right here.
So we can upload that as an additional reference along with all of these reference images of you and this prompt to help us get closer to what we want. So let's run that as well. I'm looking crazy in one of those pictures.
Let's see how this nets out here, boys. What's also interesting as well is you can use different image models. So that was all of those were ChatGPT's image model.
You can also use Google's image model. So I'm just going to generate a couple of examples here of what Google's model would look like. So you can compare and see whether or not you think Google is doing a better job of making you look more like you than OpenAI.
When you are using AI, sometimes it's good when you're just starting out to flick between a couple of different models so you learn what is best for certain things. And sometimes certain people look better with certain models and certain image models, you know, favor different things like photorealism, generating shots of like outdoor might be better on a certain model, whereas like generating shots with motion blur might be better with a different model.
There's so many like nuances. Like look, Google just did a horrific job. Google, come on, bro.
Come on, Google. Like step up the fucking game here. That's a beautiful man, but that does not look like me.
Come on, bro. So, yeah, I mean, we definitely think like open AI is king here, like they've done a much better job. But this is really interesting because this is just like one simple toggle.
You'd be like, I'd be frustrated if I was trying to generate something that looked like me and I was using something and I didn't know what the solution is. But the solution here is like you've got to try out different models. And that's why having access to an aggregator platform like OpenArt is really key.
Yeah, I mean. I mean, those latest ones don't look like me. The other ones do.
So yeah, and it could be different for other things. Like the background stuff could be better with that model versus obviously that one. So yeah, once you've got like everything that you need, the next thing, so let's say you've got your characters, you're happy with them.
You can change the outfits to be whatever you want. You could just describe the outfits and kind of generate you in any style of clothing. What we also need to do is create references of products.
If we're doing e -com or we're doing advertising, we need to dial in that product. We've got to make sure. that it looks right because we're promoting this thing.
It needs to look good. So what we did is we took a bunch of reference photos of your shoes that you're wearing right now, and we're going to do the exact same process, but we're going to do it for a product this time. So inside Claude, create a character sheet with this product or create a product sheet with all of these reference photos of the shoe here from different angles so that we can have total consistency.
I need you to be very detailed with all of the additional details, especially on like the tongue, the sole of the shoe. Make sure that like all the details remain the same. We're going to be using...
ChatGPT2 image model to create this. So tailor it towards that model so we get a more accurate prompt. And for context here as well, same to what I said before, if you can go get those product shops off the PDP page, that would be better.
Oh, dude, yeah. So like if you have the product in front of you, great, you can take loads of snaps of the product. But also if the references exist online, you could just use those too.
The most important thing is just quality of lighting. And that's like the most important thing. Really?
And for something like this as well, too, under the notion these models obviously are LLMs and do more than image and video gen. If you said this was the APL, I think it's called the Hyperloop. I'm not sure the name of it.
Like it would then search the internet and it would probably latch onto that as well, right? Yeah. I mean, like if I've uploaded this shoe in here, I think it's going to know.
It's going to find it off. Yeah. It'll probably know what that shoe is.
I mean, I could literally just ask like, you know, what is that shoe? And it'll probably be able to find out exactly what it is. But while that's loading, you know, we've got our prompt in here, which is super detailed of this shoe product sheet.
So there we go. We can go back into open art here. We can leave this reference in of the character sheet because we're going to need that.
And we can remove our old prompt that we used to create the character sheet of you and paste in this new prompt with some references of our product. So we're going to just dump all of those references in here as well. And what's interesting about like these platforms is if you like try to upload something that's like copyright infringed, maybe it's like Disney, it won't actually let you use it as a reference a lot of the time because they're trying to protect you.
Because like IP, you could get yourself in a lot of trouble. So these platforms are getting very good at making sure that you create responsibly, which is cool. And I'll show you something cool in a second as well.
So we're going to change the model back to GPT -2 so we can create an amazing prop sheet in here. And we've got my character sheet plus the references of the shoe with our crazy prompt in here. And we're just going to run this a couple of times.
So let's generate some product sheets in here. We got Turkish Mark, Egyptian Mark, Brazilian Mark, Caucasian Mark. But then we have kind of Mark.
We're kind of close. Just taking kind of a step back, all humans, all objects ideally have this specked out sheet. Yes, you need that because if all of a sudden the camera cuts to behind you and it doesn't know what it looks like, it's going to just mess up.
Same with products. Like it needs to know what the front and the back of the product looks like. These video models take these references with all of this data of you on it and it will create an incredible video.
It's context and understanding of... you know, these sheets is incredible. Like just feeding it as an ingredient, you can just build through ingredients.
So the ingredients that you need is your character references with the sheet, the product references with the sheet, and then a location reference. And that's all you need. Wow.
And the script. Yeah. And the script.
Guys, incredible free game here. Incredible free game. Jesus.
And then if there happens to be some sort of object in it and you don't like how it looks, it's probably going back to the drawing board and creating an additional object spec. And that could be like for a tree, for a table, for anything. Anything, dude.
Anything. So yeah, like in the project that I kind of just showed you, so I did like a Tamagotchi product sheet in here. So you can see we're getting like...
all sides of the product in here just so yeah so it's dialed even like what the battery looks like on the fucking inside like that's nutty so that if the character does then end up like opening up the product and we wanted to do that it will have that contextual knowledge so like in the same way if you're doing an ad on a shoe and there's you know you've got something on the inside of the tongue of the shoe and you want the character to like kind of lift up the tongue then you and show that label or that detail, it can do that reveal shot because it has the context in the character sheet.
So these character sheets are critical to, you know, the pre -production process. So here we go. So here is our product sheet, right?
Super impressive. Now, one thing AI does struggle with is sometimes is small labels, right? So if you have like a product or something and there's like a ingredients, for example, label, like those small things, Video models can really mess it up.
But this is a great example of a product which would be an amazing use case to use with AI because it's got very clear logos and like clear text in the sole. And it's big enough that it's not going to morph is what we call it. It's not going to morph when we then go to start generating videos.
So now with the references of you and these shoes, we could combine that with a location and we could create any video that we want. So yeah, to kind of talk about the project that I showed you at the start, it's the same process. So in here, we've got a character reference sheet of me.
We've also got character reference sheets of my Aussie stud right here. And we've got a reference of the product as well. So we've got everything that we need kind of dialed in.
This is absolutely incredible. So those are the only three spec sheets and then the script plus whatever you're going to show us right now? Yep.
The last thing is just a location. Right. And for locations, you can really like this is where you can use Shop Deck to help you like, you know, come up with great locations.
So if you wanted to create like this sort of like outdoorsy, woodsy environment, what we could do is if I zoom out here, if you don't mind, and just take a screenshot of all of this. So we've got all of this context in here and go back into Claude. So we can upload this for our location.
Now, location references are arguably probably one of the most important parts of the project because this locks in the color grade. Your location locks in the color grade. So if I wanted to recreate a similar shot like this, I'd just say, I'd like to create a similar shot to this where we're in the woods and there is a wooden cabin in the woods and there's like a smashed...
glass window and it's like smashed in quite like an obscure way the cabin is quite overgrown and it's like this it's like this harmony between nature and like human buildings where it's kind of like nature is overgrown around this like wooden cabin and it but it's like it's not scary it's kind of wholesome and welcoming but in like this like harmonious way so just describing something like that we can always add on the end like write out an image prompt that locks in a similar camera style, similar camera lens, so that we get the same color grade, but use my idea to create the shot that I want.
We're going to be using GPT image 2 .5 to create this image, right? And then Claude is going to now take this reference, all this context from shot deck, all of these like color palettes that we've gotten here. And boom, it's now writing out cinematic 35 millimeter film still, extreme wide establishing shot.
You know, you've got key terms, anamorphic widescreen. You know, you've got the camera body in here. You've got everything you need, the lenses, Kodak Vision, film stocks.
You've got all of this key terminology that you'd never think about if you were just going to say a cabin in the woods. You know, that's what beginners do. They think out the box.
And whereas really you need this rich context, as I always say. And this is same thing. You go right back into the program, copy and paste this.
Yes, sir. So we're going to copy that. We're going to remove everything inside here.
And then we're just going to go and create something. So here we go. So we're going to paste in that prompt.
And I'm just going to let it run. Right. So what we could also do is we could go in and we could upload this as a reference, which is the original shot from shot deck.
And we could upload that as a reference in here as well to just guide it a little bit. But here we have our location refs. So now we've gone and we've built the scene and this was without any.
guidance of a reference of an image. And these are a couple of like the different references that we've got. Now, I would say that this is still pretty AI.
Like I'm not a huge fan of this, to be honest. I would still. I wouldn't take this as my final delivery.
And if you want to optimize this, forget about the characters and the objects, which are a bit more objective. If you want to go into something like the environment, which is a bit more subjective, you see it. And let's just say you want to make the green greener and the wood a different shade.
What would you do here? Is this an LLM in OpenArt or are you going back to Claude to then re -optimize the prompt to come back into here? Right.
That's a great question. So if we wanted to like iterate from this and we were like, yeah, this is good. But it's like, yeah, as you said, this it's not like necessarily exactly what I wanted.
There's subtle things that we want to change in here. It is as simple as like going back into Claude, you know, downloading the image that we've created inside of open art here, running it up in here and just like describing what we what we do and we don't like. So in here, I quite like how the window smashed, but I don't like the size of.
This here, this feels way too out of proportion. I also feel like it's too perfect. It's just giving me like two perfect vibes.
So like we want like a bit more imperfection in here, a bit more like natural realism. So I will literally go in and say that. I'll say this is too perfect of an image.
Like I want the location to feel like I've just stumbled across it in the middle of the woods. You know, right now we have this wide establishing shot. Could we make it more of just like a bit more close in wide shot?
The house should be a bit closer to the foreground. I think the chimney is way too tall right now. It doesn't match the perspective and the proportions of the size of the building.
Any other detail that you want to change about this? Let's just see what it does and just say make the broken window half broken as opposed to like a fourth broken. I just want to see how it optimizes it.
Cool. So now Whisperflow's taken what we've said here, and it says, right, here is an edit prompt to run on this image. I've added a few of my own fixes, mainly to take out the too perfect polish.
So what we can do is we can copy this prompt, and you can see it here. It says edit this image. So what we're going to do is go...
You can just ref the same image. Yes, sir. Exactly.
So we can take the exact image that we had, which I believe is this one right here. So we can use this as a reference and then we can just paste in our prompt from Claude. I'm really excited to see what this is because this was, I'm excited to see what happens.
Yeah. And we can make all of these changes. And you can see there was only 30 seconds.
And now this looks way more realistic. Like in my opinion, this looks a bit more like real. I don't think it's done necessarily the exact changes that we wanted in terms of like the window being half broken.
But let's have a look at some of the. So that goes back to not being as prescriptive. Like if you look at how much glass is from the first one for the second one.
it was like a fourth versus a half, but we could have been like make the top half empty and the bottom half full. Yeah, it's an interesting one because what I would actually recommend is that I don't like editing images with AI. I actually always like to regenerate the image as opposed to like editing the image because every time you edit an image with AI, it effectively washes it, right?
And what happens is the picture gets more and more pixelated every time you retry to edit something. right there's a good little cheat code right here yeah yes um so what i would actually do is just remove this and go back to claude and this is like the iterative process that like i mean like nothing ever goes to plan when you're trying to do ai work so it's just a lot of back and forth and i would say like you know instead of like editing this image can you just adjust our original prompt with these changes in place so that we can just regenerate this image from scratch And then it will just rewrite our initial prompt and make those changes kind of integrated already into the prompt for us.
Guys, Uncle Rourke is giving you the cheat codes here right there. Yeah. But yeah, once you've got a location we're happy with, then you've got all the elements.
It's then just uploading those as ingredients into a video generator, which we can also do. And then just using our script, you know, that we've written to create whatever we want. And we could create like a script now and generate you in the scene wearing these shoes if we wanted to.
I mean, I feel like that would be the final piece of the puzzle, I guess, to put you in this scene wearing these shoes. And we could do like a little 15 second thing where, you know, we've got like a close up of the shoes, you like maybe walking and discovering and like looking around this place. And we can even get your voice in there as well, which is pretty crazy.
Just as long as I'm the Brazilian Mark, not the Turkish Mark. No offense to the people in Turkey. Let's talk about audio here.
So we're talking about all the kind of components. You got the character and the object specs. You got the environment, the background.
You have the script. And then what about audio? Is that like using 11 labs for me to give an input and say a sentence to get my voice?
Yeah, so there's a couple of different options when it comes to audio. Just to reiterate, we've done image generation for your references. The next thing before you go to the video stage is whether or not you need that consistent voice.
A lot of people like to just use voices that come straight out of the video model because they have like a really like great library of different voices that you can get straight out the model. However, if we really want to get your voice, there's a very weird technique that you can use to actually get your voice to show up.
And what we do. is we record you saying a sentence that phonetically pronounces every single sound in the English alphabet. And that way, basically, you can give it like eight seconds of your audio and we have everything we need to make your voice.
Another cheat code. Love that one, right? Yeah.
That's good. So I'm pretty sure it's something like the quick beige fox. There you go.
Oh my God, I got it right. That's so cool. The quick beige fox jumped over.
So you would read this sentence in your voice. in a flat tone and then we'd attach it as a reference when we go to create our project and we'd say like use this character here in this environment here wearing these shoes here and his voice sounds like x and then we'd tag the video reference that we've done which has your voice in it and then it will be able to use your voice so we will eventually get you to say this and then we'll we'll get everything dialed in cool so i have my location that i'm happy with i'm going to use This reference here of the shoes.
So let's download everything. Do you have like a character sheet that you're happier with? Do go down one.
Yep. Let me see that. Yeah, that's about close.
Let's use that. Cool. Okay.
So the next step is we've got all of those ingredients and references. What we're going to do is just inside Claude is we're going to upload all of these in here. So we have our location reference.
We have our kind of prop reference. And we have our character reference. Now, collectively, Mark and I are going to come up with a concept or a 15 second sequence of events that's going to happen inside this environment.
So it could be like a bloody commercial. It could be anything you kind of want it to be, or it could be really out there. Guys, for everybody out there listening, just for context, I have a very important meeting in about 40 minutes that I'm unprepared for.
And we actually did not prepare for this. So just so you know, this is actually live right now what we are doing. Okay.
Let's do it. So we have like 10, 15 seconds. So we have to fit a certain number of events into that time.
So you can't pack in too much. Otherwise it's going to, the AI model is going to be like, what the hell's going on? Yeah.
So let's say three main things happen in 15 seconds, right? What's going to happen? What's the first thing that happens when we go to this scene?
I think we should have me walking towards the door. Great. Yeah.
So Mark is wandering through the woods and he like approaches the front door of this wooden shack. He's looking around. He's, how are you feeling?
I'm feeling relieved that I found anything. I've been looking through the woods and I haven't found any house, any place to sleep. Yeah.
I'm relieved. Yes. So he's so relieved.
He was lost before and he's found like a shelter. So he's like. stoked but like more so relieved cool so that's like the first thing what do you say what do you say when you you you like approach the door let's go thank you god yeah okay so as he approaches the door he like kind of closes his eyes in like a relief and and says out loud thank you god cool okay that's our first beat yeah All right.
How long do we want that kind of sequence to be? Maybe like four or five -ish seconds? I was going to say four or five seconds.
Cool. Let's make this scene four to five seconds. Add in any natural cuts in there to give the, you know, the most optimal, let's say, retention for the viewer.
Great. So now the next thing. So you maybe you open the door, right?
So maybe we do like a close -up of your hand, like opening the door, and then you push it open. What happens? I don't know if we're going to need another spec for the object of me calling, but let's just say I call my wife.
Okay. On the phone or just shout like, like, how do you call your wife? I was going to do a phone, but then I feel like we would need like an iPhone.
I could just say I like orange iPhone 17 pro. It has the knowledge of what that product already looks like. Cool.
Let's do that. And then I'm going to call my wife and I'm going to say, Hey babe, I made it to shelter. Okay.
So we cut to a close -up shot of Mark opening the door and he grabs the doorknob and like pushes the door open and he walks inside. He pulls out of his pocket an orange iPhone 17 Pro Max and he calls his wife and he says, Hey, babe, I found shelter. Everything is going to be okay.
Once I'm settled, I will give you a call back. Okay. How are you saying that?
Like, what's your, how are you feeling in that moment? Let's use the same tone that I said all the other stuff with a sense of relief. Yeah.
So he's relieved. He's like, you see his like shoulders kind of come down as like, like he's no longer tense. Like he's, he's feeling like a lot more calm.
Cool. Let's just, let's just give that. Great.
Now that part is probably going to be about seven seconds, I'd say. Now we have the last part. So maybe let's do something with the shoe here because we want to get maybe some like close -up shot of the shoe.
Maybe there's like a sofa, like a rundown sofa. And you sit there and you're like proper chilling and you put your feet up and like, you know, we get a close -up of the shoe. Yeah, let's do that.
So Mark finds like this sofa that's kind of like run down, rugged, and he sits down on the sofa and there's a coffee table in there as well that's like looks overgrown. He puts his feet up on the coffee table, puts his hands behind his head and sighs in relief. Right?
Let's do it. Cool. So now with this, we're going to go back into Claude.
Copy that. And I'm going to say this. So I would like you to write me a prompt using Seedance 2 .5 to create this character wearing these shoes inside this scene.
And these are the beats of the video. And we're going to now just paste in all of like this script that we've written, right? And now what Claude is doing is it's using...
our own skill that we have created, which we'll give away to the audience, which has all of the rich language of cinematography, terminology, all of that good stuff. So it's going to write this super detailed prompt. It's going to time code everything as well.
So you can see in here, shot one, zero to zero, two seconds lost in the woods, two to three seconds. He spots it. You know, and it goes through each section and it does like slow push in, light speed ramp, 70 % speed.
And it basically maps out like he says quietly with relief. Thank you, God. And then match cut to his hand.
So it's like this editing terminology. And what's interesting here is it also adds this part at the top. And in here, it has like what we call the global look.
And there's something that I've noticed in here that it's missed. So I'm going to go and add it is if we go into shot deck right now, let's say I wanted it to feel like one of these shots, like let's say we've got the camera, like a certain camera style that we like. I mean, let's just use this one right here.
I'm going to take a screenshot of this again, and I'm just going to give it to Claude as a reference and just say, I want you to use this camera knowledge to lock in like a global look. of the video so that every shot feels like it was shot on this camera. And then it's going to go through, rewrite our prompt, and just make sure that we've got that look locked in.
Guys, we're going to put that updated prompt down below. Link in the description. Great.
So now you can see it's shot on 35mm. We've got the camera body and everything else in there. And then everything is all mapped out here.
So we can literally just go through and copy everything we need. And it's conveniently put it all into 15 seconds. So we can copy that, head back into open art, and then go into the video section, go to text to video, and then we can select the model here.
And in the same way that there's loads of different image models, there's loads of different video models too. But the best one that we want to use is Seedance 2 .5, by far the most realistic. And we're going to upload our references.
So in here, we can upload our references. So we've got the location, the product, and you. So we can upload all three of those in here.
And the last thing we need is your voice, which we'll just grab super quickly right now. So I'll just leave it here and just hit record and then read that sentence. So now the last thing we need to do is upload the voice reference.
So we've just gone and recorded that and we can go into the video section. And what's cool about Seedance is you can actually upload videos as references as well as images. And this is how you can upload the video that I showed you earlier where I'm hitting the head off of someone.
You can upload a video reference in here, you know, create character sheets of what you want to change me into, new location references, and you can change anything you want about videos as well. Limitless. Limitless, literally limitless.
So you can edit any video however you want. And guys, just for context, that black screen right there was literally him just giving me his phone, pressing record, and it's just a black screen and me speaking that sentence. Yeah.
So it's just a 10 second clip, like phone on the table, hit record, read that sentence. So now we've got all of that in here. What I am going to do just at the end is I'm just going to manually add in right at the end to say that like this character.
So what you can do with this is you can. type in at, and then you can tag specific references. So if I want to say like this character, aka at image two, which is you, and I say sounds like at video one.
So I can tag that reference to that character. So we're just going to leave that at the end and just say like, make sure the voice matches at video one. Wow.
Cool. Wow. Right.
So now we can just lastly adjust the duration of our video. You can do up to 30 seconds in one single generation. But bear in mind, a 30 second generation with Seedance 2 .5 is around $30.
It's a lot of money, right? So we're going to do 16 seconds here just to give it a little bit of extra breathing room. 1080p, which is the highest resolution that we can do right now, and 16 by 9 aspect ratio.
But we could do this for social. We could do this like 9 by 16. We could do anything we want.
So now we're just going to hit generate. And it usually takes around something around 16 seconds, might take around between 6 to 10 minutes. All right, guys, here's the moment of truth.
I'm wondering if I'm going to look like a Turkish man, a Brazilian man, a Caucasian man. I just want to reiterate, we did not plan this at all. And it is what it is right now.
So he made two different versions. Are these two different programs you use it with? All the same.
So they're both like sedans on OpenArt. So we've just run the same prompt. So it should look the same.
There'll be subtle differences. But let's have a look. Let's go.
Thank you, God. Nice hairline. Hey, babe.
I found shelter. Everything's going to be okay. Once I'm settled, I'll give you a call back.
Oh, my God. So there you go. You have a product reveal.
This is insane. It's wild. You have a product reveal in here.
Do that one again. Do it again. Do it one more time.
Let me see it one more time. Nice reveal shot.
Thank you, God.
Hey, babe. I found shelter. Everything's going to be okay.
Once I'm settled, I'll give you a call back.
That's incredible, man. Yeah. The details on the sweater, the shoes are perfect.
Yeah. It actually has my body type correct. Yeah.
Damn. So here's the next one. You got a bit of swagger on you walking around that corner.
I love this shot.
Hey, babe. I found shelter. Everything's going to be okay.
Once I'm settled, I'll give you a call back.
Tattoos on my hand, though. Yeah, so subtle difference. Like, it has, like, kind of, we call this, like, a hallucination is what we call it.
Did it pick up my other tattoo, my other small tattoo? Let's have a look. So where is that position?
I have a plus one tattoo on my hand. On your right hand? On my right hand.
Yeah, I guess you can't, I don't see it. I don't see it. Maybe on the phone scene?
No. This one, my eyes look a little weird. I think I like the first one.
I think it looks more like me on the first one. Yeah. The second one, I like the first scene better.
So you're saying that let's just say like we wouldn't, if we wanted to combine elements, we wouldn't do that. We'd start net new and just be more prescriptive. So in the AI space, I call it mining.
Right. And what you do is you mine out the best parts of the video from each of the generations and then you stitch it all together in your post -production tool. So like in Adobe Premiere Pro or whatever editing tool that you use and you take the best two seconds, three seconds from each of those generations, stitch it together until you've got the perfect piece.
So for me, I love this shot. here like the lighting the bokeh effect on here i think it's really really nice so like you know we would take the best parts of each bit and then slice it together and then you know just build upon our scenes from here you know so damn truly the opportunities are endless you can tell whatever story you want with whatever character whatever voice whatever product the opportunity is there to take advantage of it you are a legend guys I kid you not, we've been doing this for less than an hour and a half.
We're going to condense it down so you don't kind of just see the back and forth. This is absolutely incredible. I guess just the last thing, if we can just do kind of a quick summary on everything, whether it's the reference points, the character specs, like kind of just big pieces of each, how would you summarize it?
Yeah, so to summarize, you find your references of what you're inspired by. You find the color grade that you like. You write out your script.
You then formulate your idea. You start designing your characters and you start casting the characters in your scenes. You build out your locations and then we piece it all together by doing the video prompting and we generate our videos and then you edit it in post -production, add in your sound design and kind of piece it all together.
You guys just watched Rourke make an AI video from scratch. To make your first one this week, grab the free PDF down below. It has the character sheet template and the clawed skill that his internal team uses for every video.
Links at the top of the description. Enjoy the rest of the episode. I use Whisperflow for hours every single day.
I think way faster than I type. I speak about 220 words a minute and type only 45. WhisperFlow closes that gap.
It's a voice -to -text tool that works anywhere on your computer or phone. You talk and clean formatted text shows up wherever your cursor is. All right, let me show you.
I'm going to answer a real email right now without typing. You can see here Tyga is asking for feedback on our intro video for the next episode. Watch how I respond.
If you could just make a few changes, make the red a little bit brighter, the typography a little bit smaller. For the music, if you can just give me some other options, and that first photo, some more options as well. Please give me three options for each, and make sure you send it to me by end of the day Friday.
Thanks. So as you can tell, the text is way cleaner. I kind of speak in half sentences and sometimes mumble on.
This makes it very clean, formatted perfectly, and avoids all the small mistakes. This is a product that I cannot live without. I promise you that you will love it.
If you guys want to get one month free of WhisperFlow Pro, click the link below and use code OPENRESIDENCY. I'm just going to fire off tons of questions to you. Be as long -winded or as short as you want on each one of them.
Kill one myth about creative AI. It's easy and it's cheap. It's not either of those things.
Just so that the audience has context here, one video generation. of like a five second video long generation with the best AI video model right now is $5. So it's like almost a dollar a second.
So it's very, very expensive. And, you know, a lot of the time for a full project, you'll end up doing, oh, I don't know, like two, 300 generations. And some of those generations are 30 seconds long.
So the myth that it is cheap and fast and easy, I think is not necessarily true. Most valuable thing Mr. Beast taught you.
It is to have a delusional level of self -belief in yourself and believe in yourself more than anyone else will. I mean, let's just call a spade a spade. He's like the goat of creative, Mr.
Beats. So I'd love to know more. How long did you go there for?
What'd you learn? What was the intent of even going? That's a crazy story.
If you don't mind, I want to go back a little bit and just talk about how it all came around. So I was in LA. I'd just gone to an event called Google I .O.
And it's where they talk about all of the biggest AI releases that Google were releasing. And I'd attended the event and I'd come home to my Airbnb that night. And I was on a call with our team at home.
And I get a notification on Twitter that one of my idols, who's one of my inspirations, his name's Demis Hassabis, who is or was the CEO of Google DeepMind, one of the godfathers of AI, had reposted one of my videos on X. And I thought, holy fucking shit, I've made it. Like, this is the best.
day of my life. Like somebody who's inspired me so much has acknowledged my existence. It's a very validating feeling.
And in the same breath, I'm on a call with one of my colleagues and I go, holy shit, Demis Hassabis just retweeted one of my posts. I couldn't believe it. In the same breath, DM pops up on Instagram, MrBeast, hey, you free to call?
And I thought, what is going on right now? Like it was the most bizarre five minutes of my life. And we had a call and he was like, no agenda here.
I just want to kind of get to know you as a person and learn what you're doing in the AI space. And we had a 40 minute conversation. And to get 40 minutes out of that guy is hard work because.
He is a busy motherfucker. So we had this insane conversation where he just consulted for my business. And, you know, he was like, I'm five years ahead of you probably right now.
And I just, I'm going to give you all the feedback. You can take it or leave it. And he said, hey, you know, at the end of it, if you want to come out to North Carolina and come spend some time and just kick it with us, come around.
So I said, absolutely. I'm going to come and do that. So I went, I brought my lead editor and we get to North Carolina, big black.
fuck off car comes and picks you up from the airport. And I spent three days there. And it was probably the most profound three days of my life and had the biggest impact on me in my career.
I learned how businesses are operating, especially at content at scale. And it just opened to my eyes as to where AI could really fit into all this. And I was just like, I left immediately from that trip with a bunch of crazy stories.
One of the funniest ones was We did like an edit review. So where he gets all of his like 20 people in a room and he's reviewing a video that's going to go out on his main channel.
And it's got all the editors on a call. And he goes, he presses play on the video, right? Five seconds, stops, turns around, looks me dead in the eye.
And he goes, Rourke, how would you make it better? And I went, me? Like, what the fuck, me?
And he goes, yeah, how would you make it better? And I was like, maybe I'd want the video to feel more nostalgic or something. And he goes, he turns around and he goes, make the video more nostalgic.
And then he goes another five seconds, play, turn around, Rourke, how would you make it better? And we did that over and over. And it was fascinating just to see like, he wanted that like outside opinion on the project, but he cares so deeply about the craft.
Love that. TikTok, Instagram, or YouTube? Keep only one.
Instagram. And I'll say why? Because of meta ads.
Because I think like if you are building an audience, companies will pay us specifically to post on Instagram because they then can use the license on our content to boost it with meta ads if they want to, you know, and having that it's a much higher converting platform for sales than TikTok or YouTube. So at least for like meta ads or like ads in general, it's much higher ROI for businesses.
So short form there is key. Contextually, for people out there listening, he's talking about whitelisting, basically. Yes, sir.
And on the whitelisting side, you can't do that on TikTok or YouTube. You can only do it on meta, which would be Facebook or Instagram. And you're getting an extra from a bag perspective.
It's the content to get your brand and distribution. Correct. But then they're putting paid on the meta side to drive more results.
Yeah. And then they'll probably, if it's performing well, give you more money for more months. Yes.
That's a good segue into the next question. One word to describe your manager and the homie. Jake from Rakugo.
An absolute shock.
The one tool you open every single day, no exceptions. Claude. First thing you do the minute a new model drops and or you get access to it.
We're testing it to the max. So we are trying to push it. And the only way that you can know how to push it is if you know what like other AI videos already exist.
And because I've got that knowledge and that backlog of AI videos that I've consumed, I know. what people haven't done and what feels fresh and new. So we'll just like come up with crazy concepts and just see how well it interacts.
And what we're really looking for is the consistency elements of, you know, is the character moving consistently? Is the physics or the camera cuts when the camera's cutting to a different scene? Is that, you know, cohesive?
Is the timing of the videos when it's cutting to different scenes? Does it feel natural? There's all these subtle nuances you know, realism into an AI video.
So those are what we look for. AI video, how long until we genuinely can't tell? Oh, I think we're already there.
I think you are watching a lot of AI stuff that you don't know is AI. So I think we passed that point with the release of the new model, Seedance, which is from China. It is indistinguishable from reality.
Scary hours. Flexana, most you've ever made from one single piece of content? 120K.
Wildest brand request you've ever turned down. Oh my God. To make a video about AI agents for 170K.
One prompt or workflow trick you'd give away for free right now down below in the PDF. Yeah, we'll give you a clawed skill that will allow you to write. perfect prompts for your videos.
We've injected it with like two years of prompting knowledge and cinematography language. So simply by using it, you will kind of elevate your, your AI video production by like at least 30 % margin. And you can just slot it right into Claude and just ask what you want and it will write the perfect prompt for you.
What percent of quote unquote AI creators online actually know what they're doing? Three, 3%. Three or 30?
Three. Another scary one coming up here, guys. Are podcasts safe from AI?
Big time. I think podcasts are probably one of the most indefensible things because it's human sharing human value. I'm a firm believer in anything that requires human authenticity, conversation.
You don't also have to agree on people's opinions on podcasts. And I think like that's what's beautiful about them. Like people are subjective and I love that like we can challenge.
status quo. And a lot of people might not agree with like some of my takes, but I think that just by having this conversation and back and forth, this is what real is. And AI can never replace that.
Love that for us. This is a question from Joe Salvatore, a Swiss Army creative that I work with out of Lighthouse. He just wants to know, when is less is more in relation to AI?
A specific use case. Ooh. controversial take i think ai can be used in in almost any project now not to replace but to enhance and you know there are examples where let's say you're telling like the most incredible powerful story of someone climbing up a mountain it may just be that you're using ai to create like a really wicked transition shot from a to b that's it you know it's not like augmenting the the feat and the accomplishment of that person.
But it's just like using AI for VFX purposes or using AI behind the scenes to improve the footage that you already have. I actually don't think there's a project now where you shouldn't be using AI. Hot take, love it.
Last four questions that I ask everybody. Favorite book or podcast right now and why? If you guys have heard of this podcast called Open Residency.
No, I'm fucking with you. I never know what to do with that. We get that like fairly often.
I think it's just because it's the first thing that comes to mind, isn't it? Next thing I want to say, which exact episodes and what did you learn from each one of those episodes? Actually, I think the Callaway interview is one of the best.
Bro, we did part two. It's by the time this comes out. Yeah.
He's fucking nuclear, bro. He's on the limitless pill. He can articulate like what and break down to a science what it is that performs, I think, better than anyone else on the internet.
My favorite book is Harry Potter and the Goblet of Fire. I think it just escapes me from reality. Three wild takes in relation to that.
One, I've never seen Harry Potter. Two, I've never seen Lord of the Rings. And three, this man falls asleep every night to Harry Potter.
I do. You do. I don't know.
That's a wild one, bro. I thought it was like the raindrops. I go through like different phases where I have the raindrops on like a 30 minute timer.
Yeah. No, but Harry Potter is great. An entrepreneur or brand that you want to give flowers to and why?
Oh. I would say a couple. I would say Colin and Samir.
I know you've had Samir on the podcast. Legends. Yeah, man.
Like I think like I'm obsessed with the creator economy and I love this new industry that has opened up. It has quite literally changed my life and just being a creator. But then there's people who have gone and paved the way for us.
And you have to give credit to the Casey Neistats, the Mr. Beasts, the creatives who have built the businesses off the back of what they do. So I'm eternally grateful for those people for doing those things and honored that I've been able to meet them and shake their hands as well.
A creator you want to give flowers to and why? A little bit of an iteration of that. um I'm gonna go with RPM Roberto Nixon so the reason why I think he's like a creator that everyone should be inspired by is he is probably one of the most creative people I've just ever like seen on on the space he can talk about anything and people are interested and that is a very rare and unique individual so He's one person that I'm very inspired by.
A smaller creator that I'm super inspired by, there is an AI creator called Karsten who creates AI commercials. And he is just one of the best in the space. Like kudos to him.
He is a traditional creative who has just, he's underscore Karsten on Instagram. Dude creates like some of the best work in the space. So very inspired by him.
And obviously work ethic wise, I'm inspired by a creator called Hayes Sirio. I think you've met him. Yeah, good dude.
Yeah. He was very formative in the early days of my career in the AI space. He put an Instagram story up that was just saying like, thank God that my competitors have a life balance.
You know, thank God you are going for walks in the parks. And he was like, because I'm not. I'm here working hard.
When you have a life balance, I'm at home working and I'm winning. And look at that as you will. I'm very inspired by people who are dedicated to their craft.
And there was something about that that just lit a fire under me where I was like, I want to outperform and outwork everyone else. And that's all I care about. You know, that's how greatness happens.
And that's where impact comes in. So, Syrio, thank you for that. Last question.
How big can Gen HQ actually get? Yeah. So the third, I mean, this goes, obviously we have the education platform, right?
But the main thing that we've just released is our marketplace and hiring platform. This is where I see us going for nine figure exit because, and to be honest, like I will not rest until we get there, but it is a platform where. Any business can go and hire top creative AI talent right now.
These are people who have been in the space, people who come from traditional, who are interested in AI. They're impossible to find on the internet right now. They lurk in the shadows because a lot of people are scared to say that they're interested in AI.
We built GenHQ hiring and marketplace platform to be a place where top businesses from around the world, the Metas, the Googles and the likes will be hiring from our platform to find the top talent to come and work at their companies. And I believe that we will get nine figure exit in the next five years. And, you know, but the most important thing is that we achieve our mission of 100 ,000 jobs.
And that's how we can tangibly. record that because until now we couldn't really record it now we've built this hiring marketplace platform dialed in we've got numbers to hit and we're gonna do it love it i appreciate you brother
The Hook

The bait, then the rug-pull.

A 27-year-old who went broke twice now runs an eight-figure AI creative business, and his opening bet is that video editors will become the most expensive hires on any payroll. The first half explains why craft and taste still decide who wins; the second half proves it by building a product commercial live, from a ShotDeck frame to a finished Seedance clip.

CTA Breakdown

How they asked for the click.

MENTIONED ON CAMERA
Frame Gallery

Visual moments.

One-click upgrade to your Google

Get more breakdowns in your search results

Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.

Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
Watch next

More from this channel + related breakdowns.