A former self-taught editor runs DaVinci Resolve, HyperFrames and Remotion side by side for six hours to make a real product launch video, and finds out the one-shot is the easy part.
Posted
yesterday
Duration
Format
Tutorial
educational
Views
63.5K
801 likes
57 · 43
Big Idea
The argument in one line.
AI video editing works when you split the job the way a studio would: a short scripted A-cut cleaned by hand, plus AI-coded motion plates composited over it, instead of asking one model to one-shot raw footage.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A creator or small team that needs launch or explainer videos and wants to know what Claude Code can actually produce today without a human editor.
Someone choosing between HyperFrames, Remotion and the DaVinci Resolve MCP and wants a side-by-side verdict on the same footage.
An operator who keeps hearing 'this model can edit video now' on social and wants to see how many iterations a polished result really takes.
A founder with talking-head footage who wants Loom-style picture-in-picture product animations without paying for motion design.
SKIP IF…
You want a step-by-step tutorial with exact prompts and settings; this is a learning diary, not a course.
You need fully automated long-form editing of raw footage; the video shows that part still needs a human or a tightly scoped script.
TL;DR
The full version, fast.
One-shotting a launch video from five minutes of raw footage fails, but breaking video editing into its real parts makes AI surprisingly good at it. Three pipelines ran in parallel: Claude driving DaVinci Resolve through its MCP (3/10, it could not handle pauses in the A-cut), HyperFrames (7/10) and Remotion (2/10 at first, buggy). The fix was treating the work as two factories: a short scripted re-shoot cut cleanly with FireCut and a little manual blading, and motion plates of the product UI rendered by HyperFrames and Remotion, then composited picture-in-picture. Storyboards and contact sheets first, about 20 prompts of iteration, and a code-driven rebrand at the end got it to roughly 97% of a human editor's motion graphics in six hours on a Claude subscription.
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.
The skepticism about viral one-shot AI launch videos, the challenge for the day, and the 80% target against a human editor.
01:19 – 03:48
02 · What video editing actually is
Breaking editing into A-cut, B-roll, animated B-roll, music and sound design, plus the taste layer, and predicting the A-cut is where AI will fail.
03:48 – 08:14
03 · Mapping the options
Asking Grok for the lay of the land: edit real footage with DaVinci Resolve's AI control, or generate the video as code with HyperFrames or Remotion. Buying Resolve Studio and installing HyperFrames skills.
08:14 – 09:41
04 · Sponsor: Wispr Flow
Sponsor read for a voice dictation app.
09:41 – 15:13
05 · Three pipelines at once
Claude Code writes its own brief for a second session driving Resolve, then a third for Remotion. HyperFrames produces strong storyboards while Resolve struggles.
15:13 – 22:14
06 · Judging the one-shot drafts
Resolve botches the A-cut pauses (3/10), HyperFrames is remarkably good with voice (7/10), Remotion is buggy but ships a Studio editor (2/10).
22:14 – 25:00
07 · Diagnosing the failure
The brief was unfair: picking 45 good seconds out of five minutes is a scripting job, not editing. Learning one: brief the AI like you would a human.
25:00 – 29:23
08 · Two workstreams and motion plates
Splitting the job into an A-cut factory and a motion-graphics factory with a compositing step, and learning the term motion plate. Learning two: use the professional vocabulary.
29:23 – 31:26
09 · Plan before render
Contact sheets and storyboards catch problems in seconds; AI-generated UGC-style images replace fake-looking placeholders. Learning three: see the plan first.
31:26 – 34:35
10 · Script it short and re-film
Claude writes a tight script from the old transcript, it gets re-filmed, and the A-cut is made with FireCut in Resolve plus manual blading. Learning four: short video, short script.
34:35 – 37:56
11 · The human editor's cut
Reviewing Parth's manually edited version: stronger story, real-life B-roll and excellent sound design set the bar.
37:56 – 40:29
12 · Compositing and picture-in-picture
Nine plates composited over the A-cut. Both tools hit about 85%; switching to Loom-style picture-in-picture pushes Remotion to about 90%.
40:29 – 41:52
13 · Rebrand by prompt
A new logo arrives mid-project and one prompt restyles every plate. Learning five: when video is code, changing your mind is cheap.
41:52 – 43:59
14 · Final renders
Full HyperFrames and Remotion finals with music and sound effects, judged about 97% there with a few flash frames and loud music.
43:59 – 46:33
15 · Show-and-tell and verdict
About 20 prompts of iteration, six hours, no extra cost. Human plus AI is the sweet spot, and this is the worst the tools will ever be.
Atomic Insights
Lines worth screenshotting.
Asking an AI to pull a 45-second launch video out of five minutes of rambling footage is a task you would never give a human editor either.
The A-cut is where one-shot AI editing breaks first: it keeps the pauses and retakes, so everything layered on top feels disjointed.
Video editing decomposes into five jobs: clean A-cut, B-roll, animating the B-roll, music and sound design, and the taste of what to keep.
Driving DaVinci Resolve through an MCP scored about 3 out of 10 on raw footage, while HyperFrames scored about 7 on the same brief.
Remotion ships a free local Studio app, so a code-generated video arrives with a browser-based editor interface you never had to build.
A motion plate is a short, pre-rendered piece of motion graphics you drop onto a timeline, so the editor only places it instead of animating it.
Learning the professional term for what you want, like motion plate or contact sheet, makes every later prompt far more precise.
Getting the AI to show a storyboard or contact sheet before rendering turns minutes of fixing into seconds.
If you want a short video, film a short script; let the AI write the tight script from the transcript of your longer take.
Full-screen motion graphics over a talking head felt worse than picture-in-picture plates that keep the face on screen.
When the video is code, a mid-project rebrand is one prompt with the new logo instead of redoing every graphic by hand.
The polished result took about 20 rounds of short prompts, not one shot, which is the part social media demos leave out.
Six hours and no cost beyond a Claude subscription got the AI version to roughly 97% of a strong human editor's motion graphics.
The human editor's cut still won on story and sound design, which is why human plus AI is the current sweet spot.
Takeaway
Break the edit apart before you hand it over
WHAT TO LEARN
AI video tools are strong at animation and weak at editorial judgment, so the result depends on scripting tight, cutting the A-roll cleanly, and letting AI build motion plates on top.
02What video editing actually is
Video editing is five jobs: clean A-cut, B-roll, animated B-roll, music and sound, and story taste, and each one fails differently.
Expect the A-cut to be the weak point; leftover pauses and retakes make everything layered on top feel disjointed.
03Mapping the options
There are two routes: an AI operating a real editor like DaVinci Resolve, or an AI writing the video as code with HyperFrames or Remotion.
Install dependencies like FFmpeg and Whisper early, since code-based video pipelines need them for rendering and transcription.
05Three pipelines at once
You can run several AI pipelines in parallel by asking one Claude Code session to write a brief for the next one.
06Judging the one-shot drafts
On raw five-minute footage, the Resolve route scored about 3 out of 10 while HyperFrames reached about 7 on the same brief.
Remotion includes a free local Studio app, so you get a scrubbable preview interface for any code-generated video.
07Diagnosing the failure
A one-shot failure usually exposes a bad brief; asking for 45 seconds from five minutes of rambling is scripting, not editing.
Test an AI the way you would test a new hire: a scoped task with clear instructions, not the whole job at once.
08Two workstreams and motion plates
Split the work into an A-cut stream and a motion-plate stream, then use a final step that only composites the two.
A motion plate is a pre-rendered 3 to 8 second animated clip, so the editor only places it rather than animating inside the timeline.
Ask the AI what the industry term is for what you want; the right word makes every following prompt more precise.
09Plan before render
Have the AI produce a storyboard or contact sheet first, because fixing stills takes seconds and fixing renders takes much longer.
Swap flat placeholder graphics for photorealistic generated images when you need the product mockups to sell the real experience.
10Script it short and re-film
For a short video, have the AI write a tight script from your longer take's transcript and re-film it.
Silence-removal tools like FireCut give a rough first cut, but a few minutes of manual blading still finish the A-cut fastest.
11The human editor's cut
Compare against a strong human edit to see what is still missing; here it was story choices and sound design.
Real B-roll and well-timed sound effects are what make a human cut feel alive next to code-generated graphics.
12Compositing and picture-in-picture
Full-screen graphics that hide the speaker felt worse than picture-in-picture plates that keep the face visible.
Watch for AI-looking gradients and stray flash frames; naming them in a prompt usually removes them.
13Rebrand by prompt
Because every plate is code, a new logo can restyle the entire video's colors and look with one prompt.
A brand change late in a traditional edit means redoing graphics by hand, which makes code-driven video much cheaper to revise.
14Final renders
Even the strong final renders needed fixes to loud music, a few sound effects and occasional flash frames.
15Show-and-tell and verdict
Plan on around 20 rounds of short prompts for a polished result; the viral one-shot demos hide that iteration.
Six hours on an existing Claude subscription reached about 97% of a human editor's motion graphics, with the code reusable next time.
Glossary
Terms worth knowing.
A-roll
The main footage of the person talking to camera, which carries the story and audio of a video.
A-cut
The A-roll trimmed into a clean sequence with mistakes, pauses and retakes removed, before any B-roll or effects are added.
B-roll
Supporting footage layered over the A-roll, such as screen recordings, screenshots or product shots that illustrate what is being said.
J-cut and L-cut
Edits where the audio of the next or previous shot overlaps the picture of another, used to make cuts in speech feel natural instead of abrupt.
Motion plate
A short, already rendered clip of motion graphics, often a few seconds long, that is dropped onto an editing timeline as a self-contained layer.
Contact sheet
A grid of still frames or thumbnails used to review several visual directions at once before committing to a full render.
HyperFrames
An open-source framework that lets an AI write a video as HTML-based code and render every frame, with installable Claude Code skills.
Remotion
A framework for building videos in React code; it includes Remotion Studio, a local browser app for previewing and scrubbing the composition.
MCP
Model Context Protocol, a standard that lets an AI assistant connect to and control outside software, here used to drive DaVinci Resolve.
FireCut
A plugin for editing apps like Premiere and DaVinci Resolve that automatically removes silences and produces a rough first cut.
Resources
Things they pointed at.
00:00toolClaude Opus 5.5
03:48toolGrok
04:20productDaVinci Resolve Studio (about $295, AI/MCP control)
04:20toolHyperFrames
04:20toolRemotion and Remotion Studio
06:40toolClaude Code
07:00toolFFmpeg and whisper.cpp
08:14productWispr Flow (sponsor)
17:31toolEpidemic Sound
29:50toolCanva AI image generation
32:23toolFireCut for DaVinci Resolve
33:38toolWhisper Transcription app
01:00productCreatorGrid
Quotables
Lines you could clip.
24:33
“Treat the AI like a human and give it the task that you would give to a human with the same level of instruction.”
standalone rule, no setup needed→ TikTok hook↗ Tweet quote
23:50
“I don't want to be one of those people that tries to one shot something, gets slightly bad results, and then thinks AI sucks.”
relatable stance on AI skepticism→ IG reel cold open↗ Tweet quote
27:16
“Learn the words that the professionals use. As soon as I knew it was called a motion plate, I could ask the AI for exactly what I wanted.”
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.
17px
Oh my god, Opus 5 .5 is incredible at making launch videos. Look at this one shot. How far can I go trying this sort of AI editing compared to a human video editor who is very good?
If this works, I will be absolutely amazed. That is really good. Oh no, oh no.
I'm worried for video editors. We've got amazing video editors on my team. So every time a new AI model comes out, my Twitter feed fills up with people saying that this one can now edit videos.
And I've been seeing that with basically every model release for the last few months. And so I was a little bit skeptical. Now for context, I edited my own videos for about three years back when I was working as a doctor for somewhere between 10 and 40 hours a week.
And so I kind of know what a good edit feels like, but I'd never actually sat down and properly tried editing with AI. But then a public holiday came around and the team mostly had a day off and we needed a launch video for CreatorGrid, one of our platforms. Parth, one of our editors was - already doing it the normal way.
And so I decided to challenge myself for the day. Could I, as someone who's never tried AI video editing, learn how to use AI for video editing and make a video that comes as close as possible to a world -class human editor? And to be honest, if AI could get me like 80 % of the way there, I would be delighted.
This video is a little bit experimental. Usually I tell you about stuff that I already know, but in this video, I'm taking something that I don't have any idea about. And so if you want, you can watch the process of how I'm sort of learning it and figuring it out.
And hopefully there'll be something that you... you can take away from this to apply to your own workflows for learning things or for applying AI. This is gonna be a super long video, so there'll be timestamps down below.
Feel free to skip around if you like. But now before touching anything, I tried to work out what video editing actually is, because that's how we're gonna figure out how to break it down using AI. What is video editing at its core?
Okay, so if we think of when you've already got some footage. compared to when you don't. So when you've already got some footage, what is video editing?
Well, firstly, you import the footage and then you need to basically cut out all the mistakes to get a clean A cut. Then once you've got a clean A cut, you then add in B roll. It would be like the screen recordings and the screenshots and stuff.
Then the hard part, the bit that takes ages is animating the B roll so that you've got like zoom ins and zoom outs instead of just showing a screenshot, which would be boring. These launch videos often go in and out and then there's motions and transitions and fade ins and movements and swipes. pans and all those things.
So that takes a human video editor ages to do. Once you've done that, then it's like background music and sound design. So the background music, you know, we've got a particular vibe, but then sound effects like risers and swipes and swishes and all that kind of stuff.
That is ultimately what video editing comes down to. And then obviously the taste factor is... telling a good story and deciding what to cut out, deciding what to keep in.
I feel like once we break it down into all those things, these days, surely the AI should be pretty good at all that stuff. To be honest, I would be very amazed if it manages to just even get the A cut right. immediately, right?
Because when I as a human am editing a video, if there is a pause or there is a retake, you have to know where to cut and splice the footage together so that it doesn't sound weird. You know, we had a whole lesson in my course where I taught people how to use Final Cut Pro. There was a lot of time spent on just like nailing the A -cut.
Because if you can nail the A -cut, then everything becomes smoother further down the line. But if the A -cut sucks... then everything else will kind of feel a bit disjointed.
The stuff I've seen on Twitter is like super, super, super polished. They're like, oh my God, Opus 5 .5 is incredible at making launch videos. Look at this one shot.
And I'm like, there's no way that was actually one shot. Surely you had to go back and forth quite a lot and do a lot of tinkering to get that particular result. So I'm going to be very interested to see what does the one shot look like where I've just, I'm not really driving it.
I'm just like giving it the copied and pasted stuff from Grok into Claude Code and then Claude Code into Claude Code, just like Raw Dog and just see what it does. And then... When I've dabbled with AI stuff just for anything in the past, getting it to like 80 % good enough happens with one shot.
But then polishing that final 20 % takes freaking ages and the bulk of the time is spent in polishing it. So it feels good and looks good and has, you know, taste applied to it rather than just the one shot version. All right, so phase one of this project is I want to start by getting a lay of the land.
What options do we even have for AI video editing? So I'm just going to use drug grok because I don't need a VPN for it. I'm trying to figure out how to do AI video editing.
I've got some footage I need to turn into a launch video. for Creator Grid, a platform that helps teachers learn, grow and monetize. I want to see if I can use AI video editing to make a alternative edit.
Ideally, the whole thing, but also we could do a 30 to 60 second faceless launch trailer type thing using motion graphics and stuff. Or we could do it like talking head with screen shots. Like a YouTube video, some combo of all of the above.
Help me figure this out. Ah, three different pipelines. So one option is the DaVinci thing with an API.
Okay. Descript for the cut and then resolve for grade graphics and finish. Okay.
Option three. This is the Remotion Hyperframes lane. Faceless brand launch trailers.
Cool, not entirely sure what any of this means just yet, but I think we have a reasonable lay of land. All right, so quick translation because there was a lot of jargon over there. There are basically two ways AI can make you a video right now.
The first one is to edit the footage that you've actually filmed. For that, I used DaVinci Resolve, which is a proper professional video editor that recently started letting AI control it directly. But then the second is to not edit footage at all.
You instead get the AI to write the video as code and it generates every single frame. That's what tools called Hyperframes and Remotion do. And it's how a lot of those launch videos that you might've seen on YouTube and Twitter are being made.
I didn't know which approach would work. And so I decided to try both of them at the same time. Oh, what a vibe, right?
As of literally right now, Parth, one of our human video editors has actually just sent me the first draft of this video that he was manually editing. So I don't want to watch it right now, but that could be interesting to compare how far. can I go with trying this sort of AI editing stuff for the first time compared to our human video editor, who is very good.
Okay, awesome. I've got a roll filmed already. how do we get started okay firstly hagen hyperframes core skills experiment b arrow cut resolve mcp needs da vinci resolve studio 21 .1 295 all right guess i'm paying 295 for da vinci resolve studio we are buying da vinci resolve studio alrighty it worked we have just bought da vinci resolve studio for about 300 and it is an 8 .5 gigabyte file.
So it is downloading. All right. Now I need to Claude plugin marketplace.
Hey, Jen, hyperframes install. Hey, Jen, hyperframes. Let's open up Claude code.
Uh, so what I want to do is let's go local. Let's make a new folder, uh, creator grid launch video or something. And let's Opus 5 .5.
Okay. I'm trying to use hyperframes to make a launch video for creator grid. I've already got some A roll, but we don't necessarily need to use it.
Here are some notes from Grock. Let's see what Claude does. Okay, we are installing Hyperframe skills into the project.
All right. So I've already got that footage. Should I just chuck it into the project folder?
Oh, and by the way, I've added in the existing video into the project folder FYI. Ah, okay, Claude could have asked me. FFmpeg isn't installed.
Yes, by all means. please do yes install ffmpeg happy for you to install it globally on this machine pablo our co -founder has put together this clawed artifact with product views that we can use i want to see what a pure faceless just music no a voice video looks like but also i want to see what it looks like with just my voice as a voiceover and also what it looks like with both my voice and face because i think my face and voice carries trust to some degree let's do a pipeline test first it's doing stuff installing ffmpeg and whisper cpp i know what both of those things mean so i don't need to ask but if i didn't i would totally ask The screenshots are good, real UI at 4K, setting up HyperFrame's project and running the 10 -second pipeline test.
Great, it's just... doing stuff. Oh, quick note from the sponsor of this video, which is Whisperflow.
Now on a good day, I can type at about 156 words per minute, but even with that insane typing speed, I can still speak faster than I can type. And so Whisperflow is a wonderful tool for turning your voice into text in pretty much any app you might be using. So for example, I'm going to try this right now in Slack.
Hey Angus, we've got three things for Austin. Firstly, can we book the flights by Friday? Hang on, Thursday actually.
And make sure everyone's landing the evening before. Second, can you please send everyone the agenda with the planning sessions in the morning and the afternoons left free? And then third, Thirdly, can we sort out a group dinner, please, for the first night?
Ideally somewhere that we can all fit in one table. Cool, thank you so much. Okay, so that took me under 30 seconds.
And if you look at what's come out, it's turned my spiel into a very nice numbered list that makes it very easy for the person to see what I'm talking about. I made a mistake on saying Friday, no wait, Thursday, and it took that into account. Now people always say that, yeah, but my phone's got built -in dictation, to which I would say, have you seen how bad the built -in dictation on your phone is?
Let's just try it out. Hey, Tamer, are you around for a call on Saturday? No, wait, Sunday.
And can you send me the name of that book that you mentioned? So yeah. It doesn't know how to spell my brother's name.
It doesn't clock the mistake. And now if we try exactly the same thing in Whisperflow. Now, hey Tamor, are you around for a call on Saturday?
No, wait, Sunday. And can you send me the name of that book you mentioned? Great, and it even knows that this is exactly how I spell my brother's nickname.
His name is Tamor and I call him Tam and it knows exactly how to spell it. And it's automatically taken the error into account and removed it from the text, which is awesome. In my case, I've currently written 167 ,630 words on Whisperflow so far, which is a fun little stat to see.
Anyway, Whisperflow is free to use. And if you want to use it absolutely loads, then you can get Whisperflow Pro that gives you unlimited words. And you can get a month of Whisperflow Pro completely free by checking out the link in the video description.
And when they ask you how you heard about them, please do write YouTube and put my name in the, how did you hear about us box? Because then they're more likely to sponsor more of our videos, which is awesome. So thank you Whisperflow for sponsoring this video and let's get back to it.
Okay, awesome. Have we downloaded DaVinci Resolve yet? Yes, we have.
Let's install this. I've literally never worked with DaVinci Resolve, but how hard can it be? Okay.
Classic. You look at a video editor and it looks intimidating AF. What do I do here?
Help. AI? Set up AI assistance.
All right, let's do that. DaVinci has been added to these. Please restart.
Oh, wow. Okay, cool. It's already done.
So cloud desktop. Extension installer opened. DaVinci Resolve Studio.
Extension. Okay, then. Install.
Let's go. Oh, that was quick. Let's see.
What's cloud code saying now? still doing stuff it's using some tokens three percent of session limit all right we've got plenty of plenty of headroom here so what the hell do i do next on davinci i mean in theory it should be connected to cloud code right so i could just do you see the davinci resolve connection mcp i think let's see what happens yes wow we're about half an hour in so far we've managed to buy and download davinci resolve studio and claude code is cooking away on trying to do the hyperframes method.
By the way, can you write me a brief, I can put in another Claude Code chat within this project to test out DaVinci Resolve MCP for building this video as an alternative method. Nice. All right.
Well, Claude Code is basically telling me what to copy and paste into Claude Code. Boom. I love a good old copy paste from AI into itself.
Oh, wow. Okay. Cool.
Storyboards are written. No animation code yet in your bands. Gradients, glow -ins, fade -in, blah, blah, blah.
Film A, music and text, 30 seconds. This is its own edit. With no voice, the type and the product shots have to tell the whole story.
Okay, yep. Film B, your voice, 45 seconds with a 70 -second extended cut. Build it once, render it twice.
Voice only and then with your face added. Film A, silent. There's three options.
Okay, let's look at that. Oh, okay. It's fun.
All right, let's just ask these. One. demo data two yes we can show client logos three that's okay four is there a logo file um get logo from creatorgrid .io yes static no animation still render to prove the setup works please pick an a option in the b length okay so b length 45 seconds feels good and a you do two things we do everything else all right creator marketing without the chaos Nice, I quite like that.
That's pretty good. A1 option I like. Let's go.
Hey, this is so good. Okay, now let's see what Resolve's doing. Creating work folders, find Whisper models, probe source view live output, Resolve scripting API.
All right, it's just doing stuff. I guess maybe I won't even need to use. It's not visibly doing anything, but I guess it's doing stuff in Cloud Code with the DaVinci Resolve connector.
If this works, I will be absolutely amazed. There's no way it's going to work first time. There's just no way.
But this is quite cool. It's such a vibe, isn't it? Oh, open and split view.
Oh, wow. This is really cool. So at this point, we've got stuff happening, well, basically side by side.
So on this side of the screen, we've got the DaVinci Resolve method of trying to do AI video editing. On this side of the screen, we have the Cloud Code Hyperframes method of doing AI video editing. And I'm just sort of sitting back and waiting for it to do stuff, which feels...
pretty good. Yeah, this IMAX fans are starting to starting to get pretty. Yeah, this is hot.
Everything's hot. If this really works, I might just upgrade to an M5 Mac studio. Oh, wow.
Look at that. That is actually not bad. Creator marketing without the chaos.
That actually looks pretty freaking solid. That looks pretty good. I mean, as anyone can generate a still these days, but that's, you know, remains to be seen how good this is for video editing.
Let's see what happens next. All right. Both storyboards are sketched.
No animation code has been written yet. Okay, cool. Let's look at these storyboards.
It's actually remarkably good. Creator marketing and other chaos. I don't like this font, but I think this is our font.
Is that our font? Okay, I might have to redesign the creator grid branding if we do use that font. I'll get to the style later.
Pick your creators, prove the content. We handle the sourcing, negotiating, time briefing, reviewing, invoicing, paying, tracking, everything else. Nice.
Okay, so far I'm quite impressed by this storyboard. I wonder if I should use storyboards more when doing stuff. What's happening on DaVinci side?
Okay, it's still doing stuff. Okay, so DaVinci has been cooking for 20 minutes and this just seems to be running into a bunch of issues. I probably should have done this one at a time, but you know what?
Let's just see. Let's just see what it's going to do. One shot.
You know, one of the things with working with AI is figuring out what to do while it's thinking. I could try and do a third thing. Why not?
By the way. I want to try using Remotion as a third experiment running concurrently with HyperFrames and Resolve. Write a brief that I can give to a new Cloud Code instance within this project for this.
So we've got HyperFrames number one, we've got Resolve is method number two, and we've got Remotion, which is method number three. So we now have HyperFrames over here, Resolve in the middle, and then Remotion over here.
Okay, all three drafts are rendered. We have mp4 files. All right, let's see what does this look like?
And is it any good?
Okay, well, that's pretty bad as a faceless version. Let's try draft v2 voice no face. Let's see what happens.
If you're a brand trying to grow with creator marketing, you probably know that it is a total nightmare. If that resonates, I would love to tell you about creator grid where we will literally run your creator campaigns for you end to end. You don't pay anything up front.
There is no retainer. There's no subscription and you only pay for results. Entrepreneur and creator.
If we haven't met, hello. What the hell? It's Ali.
I'm a doctor turned entrepreneur and creator with about 8 million. Creatorbit is our attempt to basically give you the level of service. Oh no.
Oh no. It didn't. It screwed up the A -cut.
Just the basic A -cut. It just doesn't get it. I knew it.
I knew it. The pauses. The pauses.
The pauses are bad. It's not, it's not doing the A -cut properly. And now what if we have my face in it as well?
What does that look like? If you're a brand trying to grow with creator marketing, you probably know that it is a total nightmare. If that resonates, I would love to tell you about Creator Grid, where we will literally run your creator camp.
It means for you end to end. You don't pay anything up front. There is no retainer.
There's no subscription. And you only pay for results. Entrepreneur and creator.
If we haven't met. Hello, my name is Al. This is bad.
But this would have been magical a few years ago. Like, whoa, you could just do that with like just getting a robot to operate DaVinci Resolve. But.
I had way higher hopes of the V1. Oh, okay. Hopefully Hyperframes does a better job with this.
So I think what I may have to do is I might have to actually manually decide what to keep in in terms of what I've said. Or maybe even record a separate video that's like explicitly word for word scripted 45 seconds long. Because this one was more like five minutes long and it's just tried to figure out which bits to...
Hmm. This is frustrating. I do not like this.
Let's grab something from Epidemic Sound. I want to find a track that... people wouldn't recognize.
Oh, beautiful. Right. Okay.
So we're using Vivaldi's Spring as the background music for the Hyperframes one. Okay. Still, we've got Hyperframes doing stuff.
We've got Resolve doing stuff. And now we've got Remotion that's been doing stuff for the last 23 minutes. Also, one thing that I didn't really do is I could have, like, I've been seeing stuff on Twitter recently about that where people have come up with these like elaborate prompts.
for hyperframes. So maybe I could have just copy pasted one of those just to see what that would do. But we're not going to get ahead of ourselves.
We're just going to try and focus on the simple stuff. We have a silent draft of the hyperframes video. Shall we see what this looks like?
Okay, it's not bad. The classical music makes it kind of funny. It's a little bit janky.
It's the equivalent of if I were like 15 years old and learning video editing for the first time and deciding to use Vivaldi's Spring as the background music. It's impressive as a V1, but of course we need refinements to make it actually good. Okay, I'm going to go for lunch.
We're going to come back and my prediction is that I will be disappointed with the one shot. results of this, and then we will start the actual work of doing anything with AI, which is polishing. All righty, ladies and gentlemen, we are back after our lunch break and it is time to see what is happening.
So we're going to start with the hyperframes one. So we saw the silent one. a little bit earlier.
Let's try voice and no face. If you're a brand trying to grow with creator marketing, you probably know that it is a total nightmare. I would love to tell you about creator grid, where we will literally run your creator campaigns for you end to end.
All you have to do is basically two things. So firstly, you decide which creators you want to work with. So we'll source them for you.
You just sort of approve or reject them. We will give you a bunch of content and you just decide to approve it or not. Everything else is handled by us.
So the sourcing of creators, negotiating, reviewing the content initially, and then invoicing and paying the creators, tracking all the views, blinds, all of that annoying stuff we take off of your plate. Oh, wow. Okay.
That's like remarkably impressive. That's way better than the DaVinci Resolve one. That's really well done.
Let's see what this looks like with my face as well. If you're a brand trying to grow with creator marketing, you probably know that it is a total nightmare. I would love to tell you about CreatorGrade, where we will literally run your creator campaigns for you end to end.
All you have to do is basically two things. So firstly, you decide which creators you want to work with. So we'll source them for you and you just sort of approve or reject them.
We will give you a bunch of content and you just decide to approve it or not. Everything else is handled by us. So the sourcing of creators, negotiating, reviewing the content initially, and then invoicing and paying the creators, tracking all the views, all of that annoying stuff we take off of your plate.
Wow. Okay. That is very good.
I mean, obviously it could do with a lot of work, but that is remarkably impressive. I did not think it would be that good. Let's see how Remotion does it.
Whoa. Remotion has literally given me a video editor interface. What the hell?
I didn't realize I could do that. That's cool. Okay.
Let's see what this version looks like. If you're a brand trying to grow with creator marketing, you probably know that it is a total nightmare.
If that resonates, I would love to tell you about CreatorGrid, where we will literally run your creator campaigns for you end to end. You don't pay anything up front, there is no retainer, there's no subscription, and you only pay for results.
Okay. We need a bit of quality control on this one. I don't know why it's bugging out.
Like, it looks. I see what they're trying to do, but this is just, like, super, super, super buggy. Like, what the hell is it doing?
How the hell did you create a video editor within the browser itself? I didn't build that. The browser is Remotion Studio, which comes free with Remotion.
It's a small local web app. Interesting. Okay.
So, Remotion Studio is probably the most, like... similar to a video editing experience, but like, it was a bit buggy, just not as good as the one that Hyperframes made. But Hyperframes doesn't have like the video editing interface.
It's purely the code. And the one in DaVinci Resolve just was not very good at all. So at this point, I would rate Hyperframes as being like probably a seven out of 10.
DaVinci Resolve I would rate being probably a three out of 10. And Remotion was super buggy. So I'd probably give it a two out of 10 right now.
But it does have the right idea. It does have the right idea. Here is the problem, I think.
If we think about like the two versions of the video, if we imagine A roll is what being said, that needs to be A cut. So we go for a raw A roll. What this kind of needs is a script.
To raw A roll, then you get the A cut, and then you get the add in effects. What I was trying to get... resolve to do is basically take 10 gigabytes worth of a roll footage and hope that it would a magically be able to 45 seconds of that to turn into a cut but then also to actually be able to edit an a cut using j cuts and l cuts to the level that i would like now that is probably a little bit unfair and it's a little bit beyond what opus 5 .5 can do at this stage maybe one day you'll be able to just give it a bunch of footage and say edit this for me but today is not that day so we are going to have to at least on the Resolve front, drill down and start doing things a little bit manually, by which I mean kind of breaking things down a little bit more step by step.
Maybe that's the way to think about AI editing with Resolve. Prove the concept first for 30 seconds. You know, and okay, this was dumb of me to not realize beforehand.
In fairness, it's always good to just one shot and see what happens because then you see the holes in your prompting. Because if we were hiring a human video editor and I were testing out a brand new video editor, the way we do that is we give them 10 minutes of footage, and we ask them to just give us the first 60 seconds as an edit, because it gives us a feel for how are they doing.
It would be unfair to ask a human video editor to say, hey, of these 10 minutes, pick the best 45 seconds and turn it into a cohesive story. And then let's see the editing, because it would be a lot for a human video editor who's good at video editing, but maybe not good at scripting. Maybe actually the thing that I recorded is not suitable to be spliced into a 45 second chunk.
Yeah, I don't want to be one of those people that tries to one shot something, gets slightly sh**, and then thinks AI sucks. Like I knew it.
I knew it was using too much data. I knew the data centers were evil and et cetera, et cetera. It's like, no, I'm sure that there's definitely a way to do this.
I just need to be a little bit more sophisticated with the process and make it happen. So my learning here, learning number one is treat the AI like a human and give it the task that you would give to a human with the same level of instruction. Okay, I'm actually gonna switch out of Claude and go back into Grok.
you know what let me just use whisperflow so i think what's happening here is that we actually have two different work streams one is a pure faceless launch with video with like text and music and stuff and so that is one that doesn't need the original footage but then the second one is editing original footage those are actually two very different things editing original footage involves bringing in the footage sorting out the a cut and then doing the b -roll and the animations and stuff whereas the pure faceless is just about doing really good animations now i think what we can actually do aha is a hybrid of the two hybrid.
Because for example, if we figure out a system of nailing the editing of actual footage, like making the A -cut really good, and then separately, we had an absolutely sick way of generating insanely high quality motion video that shows the screenshots of the app with all the relevant zooms and pans and all those things.
Those are more similar to the ones that people post on Twitter. We can then combine the best of both worlds. to create a talking head explainer video that also has all of the bells and whistles that like a purely animated launch video would have.
Yes, treat them as two factories and then a third step that only composites. Stream one, a cut of real footage. Stream two, motion plates.
Hyperframes when products, screenshots, zooms, pans, connector type music hits. Build each plate as its own three to eight second composition against real UI. Drop hyperframes MP4 into the V3 as B -roll.
Talking head stays on V1. Okay, yeah, nice. Why are you calling them motion plates?
Is that like... industry terminology for this sort of thing oh it's just a post -production word for a finished piece of motion graphics that you drop into a timeline a plate is a self -contained visual layer oh i had no idea in old film work it was a painted a photograph background you composite piece people against in this workflow it's a short already rendered movie three to eight seconds of zoom pans type in product ui that does not need to be built inside resolve ah motion just means it's animated you're not asking resolve to animate the dashboard you're asking to place a clip that already has the animation in it that's clever one thing i find when learning about ai stuff when trying to do AI stuff is that I learn terminology that's actually used in the industry.
So for example, the other day I was doing some thumbnails for a thing and I learned what a contact sheet is, which is basically like a series of thumbnails for a storyboard or something like that to explore different concept directions. And then because I then had that terminology, I was able to use that terminology in future prompts.
I've never heard the phrase motion plate before, but this is exactly what I want. I want the A -cut of the video, the A -roll of me talking, and then I want really high quality motion plates to appear over the top based on what I'm saying in the video. Learning number two, learn the words that the professionals use.
Because as soon as I knew it was called a motion plate, I could then ask the AI for exactly what I wanted. All right, this is exciting. I now have more of a mental model of what I'm trying to actually do here.
Okay, so I'm going to start with what I want is I'll give... you a raw A cut, a cleaned up A cut of the video of me talking. Ultimately, what I want is really nice motion plates that I can add to the A cut using something like Resolve.
All right. Hyperframes can do this. Only the motion graphic shows with no black blocks around it.
PNG sequences for transparency, 4K output, amazing, frame rate match to your footage, one plates project with each plate as its own file. Let's just do exactly the same one for re -motion and see. Okay, I think I can basically get rid of worrying about Resolve, DaVinci Resolve for now, because I know that I don't need that for this stage of the process.
I'm now realizing that I just need to create the motion plates. Because then the script, the A -roll, the A -cut, that's like the easy part. I could always do it manually if I really wanted to.
I could chuck it into Premiere and use Firecut if I really wanted to. The A -cut is the easy part. The hard part is what I now know is called the motion plates, which is cool.
That's quite good. You pick the creators. That is all pretty good.
Wow, I like this. The only thing I don't like about it is the sort of solid colors. Okay, I like this.
I just don't like the solid colors of the backgrounds of the vertical video thumbnails. I want them to look more like actual UGC. So let's use real slash AI gen photo realistic images.
Oh wow, that's pretty good. So it's using Canva to generate casual UGC style images to replace this stuff that looks a little bit, a little bit fake. I mean, it all looks a bit fake, but I think that would sell the vision a little bit better.
Learning number three, get the AI to show you the plan before it builds anything. If you have a storyboard or a contact sheet, that takes seconds to go through and fix stuff. But if you get it to one -shot the render all the way at the end, then you have to spend a lot more time fixing stuff.
All right. So Remotion has actually done a bunch of stuff. Let's see what it looks like.
Inbox triage in 60 seconds.
You approve the content? Yep.
You pick the creators. Okay, that's really good. Your campaigns run end to end.
That is remarkably good. Well done, Remotion. That's solid.
Look at that.
Wow.
Okay, I mean, we can definitely make a nicer graph, but this is really solid. Okay, I now like Remotion again. Let me just give it the same prompts.
as I did with this one. All right, at this point it's 3 .45 PM. I have a couple of calls in about an hour.
And so I want to make as much progress as I possibly can in the next hour. I think we're getting there. Like the motion plates are actually getting surprisingly good.
So I just need a properly scripted 45 second video, which I can then just record now. Get, I don't know, one of these tools to get the A cut of it. Once I have a clean A cut that I'm happy with, I can then overlay all of these different motion plates that we're creating, whether it's from Remotion or Hyperlapse.
Okay, so at this point, I'm very happy with the images of what this stuff looks like. I think it's done a really solid job. This is the hyperframes method.
Let's see what it does when it animates it.
All right, updated reel, new views chart. Let's see what this looks like via re -motion.
Nice.
Nice. Okay, well done, re -motioned. Now at this point, the thing I'd finally clocked was that my original video was about five minutes long and I'd basically been asking the AI to dig out a 45 second launch video out of five minutes of me talking, which is a job that I would never give to a human editor either.
And so learning number four is that if you want a short video, film a short script. So I got Claude to write me a very tight script based on what it said in the original and then I filmed it again. Let's get Claude to create the script for me because it knows exactly what I said previously and it's been doing all this stuff.
In this project, we should have the transcript of the VSL I filmed for Creator Grid. Write me the V1 of the script. Okay, this is actually a very solid script.
Let's now film the thing. If you're a brand doing creative marketing, you will know that it's a total nightmare. Okay, so we have just filmed the video.
So now we need to export a clean A cut of this video. How do I most expediently export a clean A cut? You know what?
Let's download Firecut for Resolve. I think Firecut for Resolve, somewhat ironically, is actually the tool that I'm going to use for this. Firecut for DaVinci Resolve.
I've never actually used Firecut myself. Our video editors use it. And we built it as an initial tool to use in -house for our team.
Okay, so now I want to do the first cut. Boom. Can I just generate the first cut?
Let's see what happens. Nice. Okay, if you're a creator, either you pay an agency.
Yep. I'm a total noob to Resolve, having literally never used it before in my life. But let's see.
Let's see what happens. How do I zoom in and out? Oh, okay.
Yeah, fine. If you're a brand doing creator marketing, you will know that it's a total nightmare. Either you pay an agency a massive retainer and wait.
Either you pay... So Far Cut has given me a very rough first cut, but it hasn't quite... Made it.
You know what? Maybe this is just a case of me actually just doing this manually. How do I do cuts on this thing?
Command T, blade tool, boom. It's not a magnetic timeline for goodness sake. Haven't done this sort of in absolutely ages.
But this literally used to be my bread and butter in Final Cut. Produced everything else handled by us, our team and our platform. So at this point, we have just exported a clean A cut.
I probably could have done it with AI, but I didn't have the time to figure it out. I don't know, how long did it take me? So now what I want is I want the transcription.
So I'm going to use this tool, Whisper Transcription for that. Okay, we finally have a clean cut that I've just edited. This is the transcript.
I want you to generate all the motion plates for this new transcript. Boom. And let's just do the same thing over here.
So the A -cut is done in Resolve using Firecut and using a bit of manual editing. Now for the B -roll. Motion plates are being done by Hyperframes and by Remotion.
Sick. Then background music. And I think then maybe the final thing might be to add in some sound design, actually, like plops and whooshes and stuff.
Conveniently, epidemic sound has a bunch of that stuff in it. They've got swooshes, ambience, water, vehicles, whoosh. Can I just give the AI access to epidemic sound?
Because that would be handy if it could find API keys. I love it when there's API keys. So.
I don't want to do this, try and do this one shot. We're going to see what it does with just the motion plates. And then we'll add the background music and the sound effects.
But I've got it ready so that we can save some time. Bonjour.
You are on the camera. Okay, now that we've actually done this, while we're waiting, we can actually check out Parth, what Parth said. Let's see.
If you're a brand trying to grow with creator money. That's really good.
Oh, he gets it. Right. marketing you probably know that it is a total nightmare either you pay an agency a massive retainer and then you wait a few months two years for the content that may or may not be usable or you and your team is trying to like manually reach out to creators and then you're doing the outreach the negotiation the invoicing the review all of that kind of stuff yourself in -house which you just don't have the time for if that resonates i would love to tell you about creator grid where we will literally run your creator campaigns for you end to end you don't pay anything up front there is no retainer there's no subscription and you only pay oh that's good it's good Ha!
It's good. Before results. That means if you're looking for creators for UGC or content for ads, you only pay for the stuff that you approve.
Or if you launch an organic CPM campaign, you only pay for the view. I mean, yeah, he's definitely using AI systems as well. But like, it's totally fine.
Creators actually get you. If we haven't met. Hello, my name is Ali.
I'm a doctor turned entrepreneur and creator with about 8 million followers across socials. And I've been on the creator side of brand deals for the last decade. But the reason me and my team are building creator grade is that for the last two or three years, we've been trying to do creative marketing ourselves for our own software products and realizing how much of a nightmare it is.
which is why we built Creator Grid for ourselves. But then as we started speaking to other brands, everyone started saying to us, oh my goodness, I need this. And so we turned it into a product that now lots of big brands around the world are actually using for their creative marketing.
And so in the next few minutes, I want to show you exactly how it works and why I think it is a better way for you to do your creative marketing rather than working with an expensive agency or trying to do it in -house. Creator Grid is our attempt to basically give you the level of service that a done -for -you agency would.
Just way faster, way cheaper, and with a lot less bandwidth from your team. Nice. That's pretty solid team.
A creative platform is kind of like having a teammate who lives in your Slack. So you tell them what you're trying to achieve and then we will literally run the campaign for you end to end. A bit like how simple it is to run meta ads.
You set the objective. So whether you want content for ads or for you to see. Bloody hell, that's good.
See, or whether you want an organic CPM campaign and we'll handle every aspect of the delivery. All you have to do is basically two things. So firstly, you decide which creators you want to work with.
Okay. So, I mean, he's going a little bit more like real life rather than like the fake Harry Potter stuff that I was using. So we'll source them for you and you just sort of approve or reject them.
And secondly. That's clever. I like the sound design.
Oh, that's really good. Man, Parth is sick of this. Let's see if we can come close to his level of quality.
Yeah, mate, this Mac is absolutely heating up like a barbecue. Ooh, yeah. Actually, harder, harder, harder.
Hold on to it. This means that I can campaign to the wife that I need a fully decked out M5 Mac Studio, right? Which costs about 20 ,000 USD.
But it's worth it. It's my boy math. Because it will make the AI editing a little bit faster.
Even though I don't really do editing myself. I'm just doing this for the vibes. Right?
Is that okay? Can I get an M5 Mac Studio, please? We can talk about it later.
Nice. Very good. Very good.
All right. So Remotion is rendering all its things. Hyperframes is rendering all its things.
Compositing all plates over the cut into a review MP4. That's what we like to see. Okay.
I think we are done. Your first cut with all nine plates dropped in at the time codes for review only. Let's see what happens.
So this is Hyperframes. If you're a brand doing creator marketing, you will know that it's a total nightmare. Either you pay loads of money to an agency and wait ages for content, or your own team gets stuck doing everything.
Finding the creators, DMing them, negotiating with them, doing the invoices, writing briefs, contracts, like all of that stuff. CreatorGrid started as our in -house tool for dealing with all this faff, but then loads of brands told us they loved the idea, and so we spun it out. Now we run your creator campaigns end -to -end, whether it's UGC for ads, or whether you want organic views on content.
Okay, that is not bad. I wish it did more picture -in -picture. um okay not bad uh but i don't like the full screen nature of the graphics i want always picture in picture but done in a nice way okay let's see what re -motion has done If you're a brand doing creator marketing, you will know that it's a total nightmare.
Either you pay loads of money to an agency and wait ages for content, or your own team gets stuck doing everything. Finding the creators, DMing them, negotiating with them, doing the invoices, writing briefs, contracts, like all of that stuff. CreatorBit started as our in -house tool for dealing with all this faff, but then loads of brands told us they loved the idea, and so we spun it out.
Now we run your creator campaigns end -to -end, whether it's UGC for ads or whether you want organic views on content. Okay, so at this point... We have got V1 of the renders from Hyperframes and from Remotion.
Both are like pretty good. I would say they're like 85 % there. There's a little, there's a bit of polish that they could all do with.
The one, the Remotion one looks a little bit more AI sloppy with like the gradients and stuff. So maybe we can just like. Stop it from doing that.
And there were a couple of bugs, like a few like random frames and like flash frames and things in the re -motion one, but the hyper frames one looked pretty good. Okay, so I'm getting both hyper frame and re -motion to both do the picture in picture thing that sort of looks like a loom video. All right, while this is still rendering, let's do a wee wee.
I've got nine minutes until my next call. So maybe while I'm on the call. We'll get it to keep just cooking in the background for the sound effects and the background music.
Let's see if the new Remotion version is good. If you're a brand doing creative marketing, you will know that it's a total nightmare. Either you pay loads of money to an agency and wait ages for content, or your own team gets stuck doing everything.
Finding the creators, DMing them, negotiating with them, getting invoices, writing briefs, contracts, like all of that stuff. Okay, I like the Remotion one. I think we are 90 % there now.
So I'm just getting it to tweak the very first line and add a bit more juice to the motion graphics. initially. And now I have got a call, which I'm going to take on my laptop so I can keep this stuff running in the background without dying.
At this point, I was on a call with Pablo, my co -founder on CreatorGrade, and he actually sent over a brand new logo. Now, when it comes to normal video editing, doing a rebrand When you're 90 % done with a video is a total pain in the neck because it often means going back and redoing every single graphic by hand.
But because these videos were basically code, my theory was I could just give Claude the new logo and I could ask it to update the entire design system based on that logo. And so let's see how it does. So this is the new kind of logo for Creator Grid.
This is the new Creator Grid logo. Can you update the styles? etc.
of the video to match, including the relevant colors and vibe. Boom and boom. Cool.
All right. It is officially 7 .05 PM. I need to head home for family dinner and then baby bedtime with the baby girl.
But I think... Izzy is still here on some onboarding calls for her new Dream 21 community. And so until then, we're gonna see if we can wait for these draft renders to finish and then we'll see how good a job this does at just like re -skinning the videos based on the new logo.
Yeah, it's having to like re -render all of the motion plates. Learning number five, when your video is basically code, changing your mind is super cheap. Imagine how sick it would be if I had a fully decked out Mac studio that was able to run local AI models and I could just dump footage onto it and overnight it would edit the footage into what is at least a rough cut.
And so then our editors just get the rough cut that they can then finalize. That would be absolutely freaking bananas. But for now, I feel like this is a good final test of today's learning project to learn how to do AI video editing in a day.
And here we go. Boss battle. If you're a brand doing creator marketing, you will know that it's a total nightmare.
Either you pay loads of money to an agency and wait ages for content, or your own team gets stuck doing everything. Finding the creators, DMing them, negotiating with them, doing the invoices, writing briefs, contracts, like all of that stuff. CreatorGrid started as our in -house tool for dealing with all this faff, but then loads of brands told us they loved the idea, and so we spun it out.
Now we run your creator campaigns end -to -end, whether it's UGC for ads, or whether you want organic views and content. You just have to do two things. Firstly, you just approve the creators that you want to work with, that we give to you.
And secondly, you approve the content that they produce. Everything else is handled by us, our team, and our platform. There's no retainer, there's no subscription, and you only pay for the content you approve or the organic views that that content gets.
We're currently working with a bunch of big brands and we have over 10 ,000 creators on the platform just waiting to work with you. So if you'd like to try creator marketing without the chaos, please do get in touch, tell us what you're after, and we will just send you creators that might be a good fit. It costs you absolutely nothing to find out.
Nice. That is really good. Background music is a little bit egregious.
A few of the sound effects are a little bit like... There's a couple of flash frames where I got things a little bit wrong, but like, holy moly, this is like 97 % there. Okay, now this was hyperframes.
Let's check out re -motion. If you're a brand doing creator marketing, you will know that it's a total nightmare. Either you pay loads of money to an agency and wait ages for content, or your own team gets stuck doing everything.
Finding the creators, DMing them, negotiating with them, doing the invoices, writing briefs, contracts, like all of that stuff. Creative Ritz started as our in -house tool for dealing with all this faff, but then loads of brands told us they loved the idea, and so we spun it out. Now we run your creator campaigns end -to -end, whether it's UGC for ads, or whether you want organic views on content.
You just have to do two things. Firstly, you just approve the creators that you want to work with, that we give to you, and secondly, you approve the content that they produce. Everything else is handled by us, our team and our platform.
There's no retainer, there's no subscription and you only pay for the content you approve or the organic views that that content gets. We're currently working with a bunch of big brands and we have over 10 ,000 creators on the platform just waiting to work with you. So if you'd like to try creator marketing without the chaos, please do get in touch, tell us what you're after and we will just send you creators that might be a good fit.
It costs you absolutely nothing to find out. Wow, that is also not bad. That is also pretty solid.
This AI editing thing, holy moly. Do you wanna check out what I've been cooking all day? Is it filming?
It's filming. Oh, wow. These are cute.
Wait, these look quite smooth.
The music.
Luna Lovegood.
content that they produce. Everything else is handled by us, our team, and our platform. Was this one -shotted?
No, of course not. Okay. It was about 20 -shotted.
I was going to say, like, this is, like, this is really good. It's good, isn't it? It's very good.
So if we scroll, like, this is how much back and forth it was. Okay. So it's not a huge amount.
Like, and if we look at the amount of words of, you know, I'm mostly just giving it minor prompts here and there. Yeah, and it's mostly just sort of... And it's mostly just, like, doing stuff.
Doing stuff. Video editors are cooked. Except our guys, of course.
Yeah. Our guys are great. Except our guys.
Yeah. Our guys are using this a lot. I saw Pops at it.
Pops at it was also very good. It was better than this stuff. I think human plus AI is the sweet spot right now.
Now I said at the start of the day that I would be happy with the AI getting to 80 % as good as our human editors. By the end of the day, I reckon we got it to about 97%, especially on the motion graphics front. It took about six hours from start to finish and it didn't cost anything on top of my Claude subscription.
And now that I've done this once, I've firstly got all of the code that Claude generated and so it'd be easier to do it again. But more importantly, I have more of a mental model for exactly how AI video editing works. I feel like I've unlocked a part of my brain that's now...
able to think of more ways to use this miraculous technology for loads of other things in the business. Just going to be really fun. And what's flabbergasting is that this is the worst it is ever going to be.
It only gets better from here. If you've gotten to this point in the video and you enjoyed it, you might like to check out my video right over here, where I talk about tips for using AI to learn anything super fast. They'll be right over there.
Thank you so much for watching and I'll see you.
The Hook
The bait, then the rug-pull.
Every new model release brings a wave of posts claiming AI can now edit video. This is a one-day test of that claim on a real deliverable, a launch video for a live product, run against a human editor making the same video the normal way. The goal was 80% as good. The answer depended almost entirely on how the job was broken down.
Frameworks
Named ideas worth stealing.
01:19list
The five parts of video editing
Clean A-cut
Add B-roll
Animate the B-roll
Music and sound design
Taste: story and what to cut
Decomposing editing into discrete jobs reveals which ones AI handles (animation) and which it fumbles (the A-cut).
Steal forscoping any AI automation by splitting a craft into sub-jobs before prompting
25:00model
Two factories and a compositor
Stream 1: A-cut of real footage
Stream 2: motion plates rendered from code
Step 3: composite only
Keep the talking head on one track, render 3-8 second animated product plates separately, then only composite them, so no single tool has to do everything.
Steal forany talking-head explainer or launch video built with AI
24:33list
Five learnings for learning AI video
Brief the AI like a human
Learn the words professionals use
Get the AI to show the plan before it builds
For a short video, film a short script
When video is code, changing your mind is cheap
The day's lessons, each discovered by hitting a wall in the process.
Steal fora checklist when picking up any new AI-assisted craft
22:45list
Raw A-roll to finished video pipeline
Script
Raw A-roll
A-cut
Effects and motion plates
Each stage has to be right before the next; skipping the script stage forced the AI to do editorial work it could not do.
Steal forplanning any short video before recording
CTA Breakdown
How they asked for the click.
VERBAL ASK
46:15next-video
“If you've gotten to this point in the video and you enjoyed it, you might like to check out my video right over here, where I talk about tips for using AI to learn anything super fast.”
Soft end-screen handoff to a related video on learning with AI, after a sponsor read placed early at 8 minutes.
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.
A Friday inside Ali Abdaal's actual job: two student coaching calls, a newsletter written by hand, a product-marketing push, and an unusually candid conversation about what it feels like to carry a business for 14 years.
A HyperFrames builder who has only edited for six months walks through the editing vocabulary, the prompts, and the reference tricks that get Claude to cut launch videos, talking-head episodes and shorts.
A full beginner-to-advanced walkthrough of editing video with Remotion and an AI coding assistant, from your first prompt to the four-part system that makes the output reliable.
00:35
01:12
02:35
02:55
03:30
05:04
05:32
06:16
06:31
07:16
07:59
08:48
09:31
09:56
10:31
11:55
12:47
12:48
13:27
14:00
14:38
15:36
16:11
17:03
17:28
18:10
18:42
19:21
20:11
20:51
21:03
21:44
22:10
23:32
23:58
24:29
25:03
25:53
26:50
27:29
28:30
28:34
29:18
29:42
30:53
31:27
32:06
32:38
34:17
34:19
34:55
36:15
36:53
37:19
37:50
38:39
39:51
40:23
40:57
42:08
42:46
43:55
44:26
45:23
46:01
18:11
53:57
21:30
27:28
41:41
18:23