4:37:15Nick Saraev · Tutorial
The Only Codex Course You Need in 2026
A four-and-a-half hour screen share that takes four business systems from a single typed prompt all the way to a scheduled job running on somebody else's server.
August 23rdThe argument in one line.
A medium-form clips channel can now run end to end on AI, but only because every stage of the manual edit was first articulated precisely enough that an agent could repeat it without taste.
Read if. Skip if.
READ IF YOU ARE…
- You already have a backlog of long-form content (podcasts, webinars, interviews) sitting unused and want a second channel feeding off it.
- You run or manage a media team and want to know which editing stages actually survive automation and which still need a human.
- You are building agent workflows and want a worked example of a long multi-tool pipeline that ends in a published artifact, not a demo.
- You cut short-form clips for a business audience and suspect 45-second verticals aren't nurturing anyone toward a purchase.
- You want a concrete read on what a clips channel earns, with revenue and RPM comparisons rather than view counts.
SKIP IF…
- You have no long-form source material yet, since the whole pipeline assumes hours of existing recordings to mine.
- You want a plug-in tool, because the system shown here is a set of scripts, skills and prompts sold through a call, not a product you can sign up for.
- You work on Windows and need the exact stack, since the transcription step leans on Mac-only MLX Whisper and the demo runs through Raycast.
- You edit narrative or entertainment content where clip boundaries are a creative decision rather than a transcript-driven one.
The full version, fast.
Medium-form clips of three to 25 minutes nurture a buying audience in a way 45-second verticals cannot, which is why the biggest business and podcast brands all run a highlights channel. One such channel reportedly paces at $3M a year with higher RPMs than the main channel and twice its revenue on half the views, run by one person plus AI at 15 videos a day. The pipeline that reproduces it has 19 stages: onboarding, source intake, transcription, declaring the content type, clip identification, quality review, extraction, pacing, CTA assembly, SEO, thumbnails, scheduling and reporting. Agents handle most of them, but clip identification and quality review still want a human, and nothing automates until each stage is written down plainly enough to hand to a stranger.
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 →Where the time goes.
00:00 – 02:30
01 · Cold open
The claim: a medium-form clips channel automated end to end, no human in the loop, capped only by token spend. Most people talking about this online have never shipped it.
02:30 – 04:10
02 · Why medium-form
Long-form, short-form and the gap between them. Three to 25 minute clips nurture a qualified audience in a way 30 to 60 second verticals cannot, and every major podcast already runs one.
04:10 – 05:17
03 · Articulate, then automate
The governing rule, borrowed from a mentor: you cannot automate what you cannot articulate. Years of doing the edit by hand are what made the written spec possible.
05:17 – 06:34
04 · DIY, DWY or DFY
An upfront disclosure that the system is for sale three ways, with the free video serving as the proof.
06:34 – 08:37
05 · Process map and the revenue claim
A 19-stage map built from the workflow, then the source email: one person plus AI shipping 15+ videos a day, pacing $3M a year, higher RPM than the main channel, twice its revenue on half the views.
08:37 – 10:09
06 · Three clips in 35 minutes
A live receipt. One prompt at 11:15, three source videos (Buffett, Jobs, a podcast episode) turned into three scheduled public uploads 35 minutes later. Six would have cost almost no extra time.
10:09 – 12:14
07 · Onboarding and source intake
Verifying the channel to unlock links, API scheduling and A/B testing, then getting source material into one consistent folder on Drive or Frame.io rather than scattered across platforms.
12:14 – 14:36
08 · Transcription and content type
Transcribe in Premiere, DaVinci, CapCut or locally with MLX Whisper, then tell the agent what kind of recording it is, because expected clip length depends on the format.
14:36 – 20:59
09 · Finding a clip by hand
The long demonstration. Scrubbing a Steve Jobs interview in Premiere at double speed, catching the cold-call story about Bill Hewlett, setting in and out points, color-coding the segment and leaving a named marker.
20:59 – 22:50
10 · Quality check and extraction
Read the first and last 30 seconds of each candidate to catch bad boundaries, then lift the approved range into a clips sequence. The system does this with scripted windows and FFmpeg cuts.
22:50 – 24:26
11 · The used-clips log
A log of what has already been published so a second pass over the same source does not duplicate a clip. Pure coordination overhead that a solo human never needed.
24:26 – 26:01
12 · Pacing and CTA placement
Strip pauses and filler words with Premiere's transcript filter, then stitch a CTA either at the end or a third of the way in, dropped right before a payoff line like a commercial break.
26:01 – 30:42
13 · Transcripts, SEO and Notion
Export the clip transcript, run it through an Optimize YouTube SEO skill for titles, description, tags and hashtags, and have the agent write the result into a Notion card the scheduler can read later.
30:42 – 40:15
14 · Building a thumbnail by hand
A full Photoshop pass at 1280x720: grab a frame, generative upscale the face, extend the background, add a quote, gradient, drop shadow and a camera raw grade.
40:15 – 44:19
15 · AI thumbnails in one prompt
Feed the finished thumbnail back as a reference with a new face and a new quote. One detailed prompt reproduces the hour of manual work, and the system runs this as a thumbnail factory.
44:19 – 46:29
16 · Scheduling in YouTube Studio
Upload, paste the stored title and description, add the thumbnail (needs phone verification), set playlist, audience, tags, end screens, clear the copyright check and schedule.
46:29 – 47:43
17 · Manual versus automated
The direct comparison: a little over an hour for one clip by hand, 35 minutes for three by prompt. Plus why upload and A/B testing keep a human, and how reporting gets pushed to Slack.
47:43 – 49:31
18 · Close and offer
A six-month client channel shown as proof, then the three ways to buy, a request for comments on what needs explaining better, and a promise of a V2.
Lines worth screenshotting.
- You cannot automate what you cannot articulate, so the bottleneck in an AI pipeline is almost never the model, it is the undocumented step living in someone's head.
- A highlights channel pulled half the views of the main channel and twice the revenue, because medium-form viewers are self-selected buyers rather than scrollers.
- Short-form clips buy discoverability but do almost nothing to nurture a viewer toward a purchase, which is the job medium-form does.
- Three source videos went from raw files to three published YouTube clips in 35 minutes, and six would have taken roughly the same time.
- Transcription and clip identification are the only stages that scale with video length, so doubling the output costs almost no extra wall-clock time.
- The same job done by hand in Premiere and Photoshop took a little over an hour for a single clip.
- An agent has to be told the content type (Q&A, panel, podcast, interview) because expected clip length depends on it: coaching exchanges run three to 10 minutes, podcast through lines eight to 25.
- Reading the whole transcript in one pass beats splitting it across a swarm of agents, because a clip's value depends on context a chunked reader never sees.
- Checking only the first and last 30 seconds of a candidate clip catches most bad boundaries without rewatching anything.
- A used-clips log is pure automation overhead: a human editor remembers what they already cut, an agent and a distributed team do not.
- Placing a CTA a third of the way into a clip reaches far more viewers than one at the end, because retention is highest early.
- The right spot for that CTA is immediately before a payoff line, so the viewer sits through it the way they sit through a commercial break.
- Verifying the channel with a phone number unlocks third-party links, API scheduling to public, and A/B title and thumbnail testing.
- The YouTube API still cannot run A/B title or thumbnail tests, which is why the upload stage keeps a human in the loop.
- Thumbnails are optional on medium-form clips, and the reference channel ran without them for months.
- One clearly written thumbnail prompt plus a reference image reproduced an hour of Photoshop work in a single generation.
- A strict naming convention and a single fixed export folder save tokens, because an agent that hunts for files burns context on every run.
- Centralizing titles, descriptions, tags and thumbnails in Notion gives the scheduling agent one place to read from instead of a scavenger hunt.
- Doing the work manually for years is what made the automation possible, because every rule the agent follows was first a habit.
- Audience quality, not volume, is what makes a clips channel earn: problem-specific videos attract viewers already looking for a solution.
Write the process down before you automate it
WHAT TO LEARN
The pipeline works because every judgement an editor used to make silently was first written out as a rule, which is also why the two stages that resist writing down still keep a human.
02Why medium-form
- Medium-form clips of three to 25 minutes nurture a buying audience, while 30 to 60 second verticals mostly buy discoverability.
- Every major podcast and business brand already runs a clips channel, which makes this a proven format rather than an experiment.
- A highlights channel pulled half the main channel's views and twice its revenue, because problem-specific videos attract viewers already looking for an answer.
03Articulate, then automate
- Automation is downstream of documentation: if a step only exists in someone's head, an agent will guess at it and get it wrong.
- Doing the work manually for years is what made the spec writable, so repetition is research rather than wasted time.
05Process map and the revenue claim
- Reported scale on the reference channel is 15 or more videos a day from one person plus AI, pacing at roughly $3M a year with higher RPMs than the main channel.
- The limit on output is source material and token spend, not labor, which inverts how a media operation is usually staffed.
06Three clips in 35 minutes
- Three source videos became three published clips in 35 minutes, and six would have cost roughly the same time since only transcription and clip-finding scale with volume.
07Onboarding and source intake
- Verifying a channel with a phone number unlocks third-party description links, API scheduling to public, and A/B title and thumbnail testing.
- Keep every source file in one consistent folder, because an agent that has to hunt for the new location burns context on every run.
08Transcription and content type
- Transcription is the real gate on an automated pipeline, and a local model removes the per-minute API bill entirely.
- Tell the agent what kind of recording it is, since a coaching exchange runs three to 10 minutes while a podcast through line runs eight to 25.
09Finding a clip by hand
- An agent reads the subtitle file instead of watching, so a clip is found by locating a complete through line in text with start and end timecodes.
- Have the model read the whole transcript in one pass rather than chunking it across agents, because a clip's worth depends on context a chunked reader never sees.
- Clip identification is the stage models are still worst at, so review their picks rather than publishing them unseen.
10Quality check and extraction
- Checking only the first and last 30 seconds of a candidate catches most bad boundaries without rewatching the source.
11The used-clips log
- Keep a log of which segments have already been published, since a human editor remembers and a fresh agent run does not.
12Pacing and CTA placement
- Strip pauses and filler words from a clip before publishing, since good pacing is what separates a cut from a raw excerpt.
- Put the call to action about a third of the way in rather than at the end, because far more viewers reach 30% than 100%.
- Drop that CTA immediately before a payoff line, so the viewer waits through it the way they wait through a commercial break.
13Transcripts, SEO and Notion
- Generate the title, description, tags and hashtags from the clip transcript, since the clip already contains the search language.
- Store every published asset in one database the scheduling step reads from, so no stage has to go hunting for the last stage's output.
15AI thumbnails in one prompt
- Thumbnails are optional on medium-form clips, and the reference channel ran for months without them.
- One detailed prompt plus a finished thumbnail as a reference reproduced an hour of manual design work, so the first one you make by hand becomes the template for every one after.
- A strict naming convention is a token-saving measure once agents are involved, not just tidiness.
17Manual versus automated
- Scheduling still keeps a human because the API cannot run A/B title or thumbnail tests, which is where easy performance gains sit.
- Push channel analytics into wherever your team already reads, so the pipeline closes the loop instead of ending at publish.
Terms worth knowing.
- Medium-form clip
- A three to 25 minute excerpt pulled from a long recording, long enough to carry a complete idea. It sits between a full podcast episode and a 45-second vertical clip.
- Highlights channel
- A second YouTube channel that posts nothing but clips cut from a main channel's long-form library, used to reach search and recommendation traffic the main channel misses.
- RPM
- Revenue per mille, the ad earnings a channel makes per thousand views. A higher RPM means the same view count is worth more, usually because the audience is more commercially valuable.
- Through line
- A continuous thread of argument or story inside a long conversation that can stand alone once cut out. Finding one is what separates a clip from an arbitrary excerpt.
- SRT
- A plain-text subtitle file that pairs lines of speech with start and end timecodes. Agents read it instead of watching video, since timecodes are what a cut actually needs.
- Single-pass read
- Handing an agent an entire transcript in one context window rather than splitting it into chunks across several agents. It costs more tokens but preserves the context that makes a clip coherent.
- Used-clips log
- A running file recording which segments of each source have already been published, so a later pass does not cut and post the same moment twice.
- MLX Whisper
- A port of OpenAI's Whisper speech-to-text model optimized for Apple Silicon, which transcribes long recordings locally without an API bill.
- FFmpeg
- A free command-line tool that cuts, converts and re-encodes video. It is what lets a script perform edits that would otherwise require opening an editor.
- A/B title testing
- A YouTube feature that rotates up to three titles or thumbnails on one video and keeps the best performer. It unlocks only on verified channels and cannot be triggered through the API.
- DIY, DWY, DFY
- Three delivery tiers for the same system: do it yourself (you get the files), done with you (the seller implements it and trains your team), done for you (an agency runs the channel).
- Generative upscale
- An AI enlargement of a low-resolution image that invents plausible detail instead of simply stretching pixels. Useful for turning a video frame into a thumbnail-sized face.
Things they pointed at.
06:57linkAlex Hormozi's AI highlights channel email
04:10channelNicolas Cole
48:20channelDaniel Fazio client highlights channel
Lines you could clip.
04:10
“You cannot automate what you can't articulate.”
07:35
“Our highlight channel generated twice the revenue of our main channel.”
09:35
“Thirty-five minutes later, it had gone from just receiving brand new videos to posting all of them.”
46:30
“I've been here for a little over an hour now. So as you can see, this is how long the process would take if you were to do it manually, clip by clip.”
02:50
“These clips aren't the most valuable tool to help nurture your audience. Sure, they might do a decent job at discoverability, but they're not the best for making your audience stick around.”
24:55
“You want to place it during a cliffhanger moment, sort of like every commercial break on television ever does it.”
01:35
“The people who have actually done this stuff are the slop factory kings. The people who are actually creating valuable content, most of them weren't able to automate this yet.”
02:08
“I guess I'm just helping YouTube maximize user screen time. But let's not worry about that, because that's not what we're here for.”
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
I did it. I cut my hair. No, that's not what I did.
What I did was I was able to automate a medium form clips channel from start to finish the whole thing end to end using AI. It needs no human in the loop and it's able to pump out as many videos as you want. per day fully using AI.
You're only limited by the amount of tokens you can spend in a day. Now, why does this matter? And there are obviously some caveats, what I just said.
But what's really cool is that up until this point, you've probably seen a lot of people on Twitter talking about automating a clips channel from start to finish. But a lot of people talking about it have only spoken about it in theory, like they can conceptualize it. Sure.
But I would guess that like 99 .9 % of them, there's a big spider there. I would guess that 99 .9 % of them haven't ever done it themselves. The people who have actually done this stuff are so the guys who are creating like these massive like networks of channels and channels that are just like pumping out tons of slop AI content on the internet.
Those guys have it figured out. But the people who are actually creating valuable content, content that's actually relevant to the audience, most of them haven't, weren't able to automate this full thing with AI yet. And I set out to do it.
It took me, it's safe to say it took me like 100 hours -ish to do it, but I was able to do so. I started working at it a couple months ago, actually. And then now I set out, ever since Astra released, I was like, okay, there's so much more we can probably be doing with Astra and Fable.
And so I was like, okay, let me try and actually sit down and... figured this out and i was able to do so and what's really cool about this again is that what we're creating is relevant content content that's not like pure slop and content that an audience of valuable buyers like a qualified audience might actually find interesting or will actually find interesting so this is great because you're not just pumping out slop you're creating relevant valuable content that people can consume and so what you're essentially doing is you're giving youtube the power to feed people super relevant specific content for them so they can consume a lot of content and just sit around and watch YouTube all day because they have the most specific, highest relevant content that was tailored for them, being served to them at all times.
So yeah, I guess . But let's not worry about that because that's not what we're here for. If I'm helping create the dystopia that we live in and the apocalypse that might ensue or that might come in the near future.
That's all right. Let's just focus on what we have. Let's just try and be optimistic about it, right?
But what's really cool about this, again, I'm working on a different video, which is entirely focused on explaining the opportunity and the mechanism behind medium form clips and why they're so valuable. So if it's already out, you'll probably see it somewhere in the screen. I'll put it in the description as well.
But just to give you a brief overview of what we're talking about is in the current content landscape, we have super long form stuff. So we have long form YouTube videos, we have podcasts, we have webinars, events, all that stuff. And we have super short form clips or not super short form, but we have short form clips.
which are 30 to 45 to 60 second long clips that get chopped out of these long form conversations or these long form sources and posted across like the vertical platforms. So Instagram reels, YouTube shorts, TikTok and so forth. But the problem is these clips aren't the most valuable tool to help nurture your audience or to improve your discoverability.
Sure, they might do a decent job at that, but they're not the best for making your audience stick around or again, just nurture the audience and nurture the kind of relationship you want with a viewer that might be interested in buying your product or service. I'm talking. about this through a lens of business content or content that helps you generate a certain result i think this is the best opportunity or the best way to leverage this opportunity but even if you're like a media company if you just have a podcast or something like that it's easy to notice that all the biggest podcasts have some sort of medium form mechanism running and a lot of them have had it for a long time you look at joe rogan you look at lex friedman they have clips channels which bring in like millions and millions and millions of views every single day every single month for them which obviously helps with discoverability and helps them establish and strengthen their brand.
And so this is why this is such a cool opportunity. And what's even cooler, again, going back to the business realm of things or the business niche, Alex Ramosi has been running his Highlights channel for a few months now. And he recently sent out an email talking about how they run it entirely with AI, which I'm going to show you in a second.
And he also said it in a video that their Highlights channel is generating a ton of revenue. I'll also show that to you in a second. But yeah, I just wanted to introduce the concept and talk about it for a little bit.
This is going to be a long video, so I figured you would have an extra five minutes to just sit through. a bit of a longer introduction. But yeah, my whole thesis behind this is the following.
So I was able to automate this whole thing because I was doing this whole thing manually for a long time. I've been working with content for many years now. And so I understand this very deeply.
I grasp everything that goes into this process. And to quote something that Nicholas Koh, who I consider a mentor, has said many times, you cannot automate what you can't articulate. And so for you to be able to automate a process from start to finish, you have to understand how that process works and what goes into it from like the very first step.
to the very last step. And you have to be able to clearly and very objectively articulate each step of the process so that you can then instruct an AI or for this matter, like a different person, if you were just to like hand this off to someone else so they can repeat the process themselves. And so this is why I think I was able to successfully do it is because I have done it many times manually in the past.
I was doing it assisted by AI for a long time too, which I still think might be the best way to use this or the best way to go about this. So I'm going to show you how to do it fully manually. So you can then just send this to your media team or send this to the person on your team.
responsible for content and they can learn it, they can understand it. And so they can try and improve or like streamline their current process, or they can simply watch all of this and try and automate it themselves. And to be very transparent from the get -go, the purpose of this video is also to help you implement something like this in your business.
So whether you just want to buy my workflows and the systems and everything that I've put together, so you can just like implement this into your business and run it yourself. So the DIY option of this is available for sale. You can click the first link in the description.
You can book a call. The whole system is available for sale. If you want the DWY version of the system, so if you want me to help implement this into your business.
If you already have an existing media department and someone who's responsible for your content and just want to empower them and make them more efficient and streamline their processes, I can also help you with that. I can also sell you the consulting on that. Or if you have a large enough backlog of content and you just want someone to run this for you entirely like done for you as a service, that's exactly what my agency does for clients.
And so whatever option best fits your current circumstance, you can just click the first link in the description to book a call with me and we can talk about it and figure something out for you. Otherwise, if you're just hearing this video because you want to learn more, about it or you want to you yourself provide this as a service for clients and want to better understand how to go about this whole thing, then welcome.
This video is free and I hope that you can get some value from this video so you can make your own processes more efficient or maybe learn something that you weren't doing before. Anyhow, let's get into it. I'm going to move to a different studio right now because there I have a bigger monitor so I can walk you through the whole thing.
The full masterclass on how to create a highlights channel from start to finish with AI, how to do it manually, and then how to automate the whole thing with AI. So let's do it. Boom, we're here.
So let's do it. So this is the first thing I want to show you all. I mapped everything that we're going to talk about in this masterclass, in this page here that lives in vba .io forward slash masterclass map.
So if you want to access that, I'll put the link to that in the description too. This literally is just Claude turning every one of the steps that we're going to cover in the masterclass are just mapped out in this page here. So you can go through it, you can look through it yourself, but yeah, let's get started.
So the first thing I want to show you guys is the Hormozy that email sent, is the email that Hormozy sent a month ago, about a month ago. Here's the subject line, my AI. YouTube channel.
We're using AI in the pre -post for my main channels, but we're using it entirely end -to -end in my highlights YouTube channel, yada, yada. Here, he later says that there's one person who's putting out 100 % of the volume. So it's not fully AI.
There's a person who's managing the AI naturally. So heavily assisted by AI to get 15 plus videos out per day. Given the volume of raw content and like source material that Hormozy has, I'm even surprised that they're not posting more.
So like they could be posting a video an hour. They could be posting two videos every hour with the proper AI workflows, with the proper AI systems. So maybe They're just choosing to not pump out as much volume as they could.
Or maybe this person isn't using the AI to its fullest. This is not an insult to them, but by any means, it's just, I'm going to show you guys how what I'm talking about here allows us to create even more than that per day. And so that's really cool.
And what it's producing, it's already pacing $3 million per year, which I expect to grow. So keep optimizing it. And it has higher RPMs than my main channel.
And obviously this is because this, like the viewer that is watching the Hermosi highlights channel or the medium form stuff from Hermosi is a much more qualified viewer. It's a much more qualified audience because those videos, are all very problem specific.
Those viewers are all very like focused on business, tactical business stuff. So here's the cool part. It takes zero extra time.
It goes to show how small views can create. And so I don't know if I'm allowed to show this video or not. If not, Daddy Hermosi, please do not sue me.
This is just going to be a short clip that I got from his course thing. Based on the attributable data that we have. All right.
So there's attributions very tough with organic, but I'll say just directional data. Our highlight channel generated twice the revenue of our main channel.
So this is from five months ago, as you can see here, it's from April of 2026. And he had just said in the video that the highlights channel had actually pulled like half the views when compared to the main channel. And so half the views and double the revenue, which is pretty fucking cool.
And so, yeah, just to show you guys an example of how the fully automated thing can work or like how the fully automated thing works. I had Codex, let me pull it up here. I had Mr.
Codex here run the system. So I had done a test run before starting, but I ran the system a second time and I want to pull videos from this test run, this folder that I put together here. And what it has, what it had was a video, a lesson from Warren Buffett, the last interview from Steve Jobs, and then my first million podcast with Rory Sutherland.
And as you can see, I gave it a prompt at 1115. I said to schedule it all to YouTube to go live at 11 a .m. Eastern.
And then it just asked me something unrelated. And there we go. 35 minutes later.
Yeah, 35 minutes later, it had posted. It had gone from just receiving the videos, brand new, fresh new videos to posting all of them. As you can see them here to this dummy YouTube channel I created.
So we have, let me just refresh this page. They're all public. Now, as you can see, this is a nine minute clip.
This is a nine minute clip. It's a 10 minute clip from these source videos. So again, 35 minutes it was able to post three videos in a snap and since the longest part of the process is actually getting the transcripts for the videos and identifying identifying the clips if we wanted to post six videos instead of three it wouldn't have taken us any more time so this is really cool and this just helps demonstrate how you could create more than 15 videos per day if you wanted with a system like this it's just a matter of how much source material do you have which is really cool and so i'm going to go through this document here with you guys and then i'm going to set out to explain how to do everything manually and then this the document itself already like says how we're doing it with the system how we're automating the whole thing but i'll just reread it for everyone so the first thing the channel onboarding this is just this is something that will always have to involve a human in the loop obviously you can automate it with computer use but if you want to do it yourself this takes five minutes and it's just easier so basically what this means is creating a youtube channel and verifying it the reason why i want to verify it with the phone number and with your identity is because that unlocks a whole new set of youtube features such as including third -party links in the description scheduling videos via api and allowing them to go public title testing
and thumbnail testing, ABC title testing and thumbnail testing, all of those only get unlocked when you verify the channel. So you have to do that. What the system actually does is it helps you onboard you through the whole thing.
So we have the skill that onboards, if you were to buy the DIY version of this, if you were to implement this yourself, you would have an agent onboard you on the whole thing, give you the links to everything, show you what to verify, how to verify and connect everything for you automatically. So the second thing then is the source intake.
And this can happen in a couple of different ways or in a few different ways. But one of them is say you're creating long form content or you're creating a podcast or you're creating a YouTube long form or you're going on interviews, whatever, these all live somewhere.
So they might be available publicly. So you can just, someone might have to go through your, look you up on YouTube or look you up in Google and find the stuff that find the long form videos or the long form appearances you've had, download them and store them somewhere. Or this can already be automatically fed to a drive folder that you currently have in your media operation.
Again, this depends on where you are in the spectrum of, I already have a media operation running and I have processes and stuff. or I have zero. But basically what you want to do here, as I said, is you want to include as much source material as you can in a drive or frame IO folder so that the person who's responsible, the human who's responsible for editing it can pull from there consistently.
Again, even the uploading process to these folders can be automated nowadays. What you can also do is you can download them automatically, which is something that the system does. And actually what's really cool is that frame IO has recently come out with a new tool that allows you to interact with videos.
And so edit videos and stuff like that without even having to download the files in a local So maybe this is a potential upgrade for the system because we now don't even have to have stuff downloaded. Everything might be able to be operated from the cloud, which is also pretty fucking cool.
But yeah. download again this is just how this happens automatically with the system and then the transcriptions if you want to transcribe everything automatically this is where i'm going to start pulling some tools here but you could do it in premiere they have an auto transcribe tool you can do it in davinci if i'm not mistaken you can obviously do it in cap cut i use the adobe creative cloud suite so everything that i use is from adobe i use photoshop i use premiere and that's what i'm going to be walking you guys through and every tool that i mentioned in this video if there's any third -party tool or something else that i might use i'll leave links to all of it in the description below and i think the the transcript or the description here will also do a good job at listing out all the different tools that we use.
So for example, here, this is all proprietary stuff. This is the skills and the scripts that we use to find clips and to make the whole system run. But this, for example, YouTube DLP is something that you can find online and you can just have Claude or CatchVT install into your machine.
And so again, going back to the transcription step of things, if you have the system, the automated system running, what you would use if you... run on a Mac, you would use MLX Whisper Turbo, which is currently the best transcription model or like AI transcription model for videos that live in your machine. So basically what this does is it runs on the background.
You have a video of an hour, two hours, or however long the video is. And MLX Whisper Turbo just transcribes the whole thing for you. And with the system, it stores it within the workspace where everything lives and everything communicates with itself.
So now we move on to fighting the eclipse. So this is steps four to eight. First one is just declaring the content type.
This is something that's a bit specific to the system because it needs this level of specificity for it work for Dwight. understand what it's working with.
If it was a human editor, like if you were handing this off to a human editor who has taste, who understands what he's doing, this would be very like intuitive. You don't have to think, oh, I'm editing a long form conversation versus I'm editing an interview versus I'm editing a Q &A session versus I'm editing an event panel.
Again, this just comes naturally to us, but AI needs this sort of specificity, these content types separated by like, this is what you're working with. So it can properly understand what to look for and what clips to extract from each source video. So again, this is just, it's also included here, but the agent doesn't guess this.
which is what a human would do because the expected clip length depends on it. A coaching exchange runs three to 10 minutes. This would be like a Q &A example type.
A podcast through line can run from eight to 25 minutes long. So like a clip from a podcast can be much longer than a question or just simple Q &A in a panel or in a podcast, for example. So this is just part of what we can do.
Like the agent can decipher this for itself, but I also chose to include like a human in the loop here because what I think this helps with is just, it just gives the agent as much certainty as possible. Like this is a Q &A. This folder only has Q &A.
it's in it. This folder only has long -form conversations, and so it knows exactly what it's working with. Number five, which also is automated, but if you were to do it manually, clip identification.
So we have reading the transcript. Obviously, a human wouldn't read the transcript. They would just watch the video.
It's just because agents cannot watch videos at this current moment. Looking for through lines. So this is just literally just going through a long -form video and looking for a clip.
What this looks like with the system is that the agent reads the SRT, which is the transcript file of the video, with a criteria file in a single pass. So what this means in practice is it's not chunking the view it's not splitting it up which is what a swarm of agents would do to preserve token usage and so it's looking for the best possible clips in the transcript and this is something that i would say is the meat of the whole system because it's very important that you have very specific guidelines or very specific descriptions of what makes a good clip i think this is already a good segue for us to open premiere and start doing some of this stuff manually to show how this would actually work in practice so let me pull up premiere here and show you how i organize everything within my own machine so here we have premiere I'm going to create a new project here.
I'm going to create a new folder here called Medium Form Masterclass. I'm just going to do this source and clips as the Premiere file. I'm going to create it here.
Let me bring Premiere back to the main screen. I'm going to go to edit, drag the project here, and I'm just going to start organizing myself. I think it's really done.
Just change the work. workspace settings here too. I think it's really important for you to properly organize yourself.
I think the more organized you are, the better you can navigate in your own workflow, the more efficient you'll tend to be. And so I just want to show you guys how I do all of it. I'm just going to create a sequence and take a 1080p sequence here.
I'm just going to rename this to sources and then create a sequence out of it. And then a new one called clips. I'm going to create a folder for the raw files.
And then I'm going to pull these videos here that I had just mentioned. to the raw files and then i'm going to drag them all to the sources sequence what we could do is we could create us an individual sequence for each one of the clips but i don't think that will be necessary for the sake of this example i'm not going to do that just because we can walk them we can walk through all of them in a single sequence here which is called sources so basically what we would do right now is we would manually if we were to do the whole process manually is we would sit through this recording here so 16 years where the Clip starts.
It's the interview with Steve Jobs. Let me just adjust the scale here so he looks a bit larger.
And great tip if you want to go through a podcast or a long -form source faster and find clips, identify clips, is you can just press L on the keyboard, and that will speed up the video in Premiere. And if you press L twice, it will speed it up even more.
Basically, this is now just going to be him telling stories and being interviewed. So what we want to do here is find a worthy clip for us to post on YouTube.
Even with that, it was still remarkable, especially for a 10 -year -old, that you could write a program in BASIC, let's say, or FORTRAN. And actually, this machine... And then I was part of this group at Hewlett -Packard.
When I was 12, I called up Bill Hewlett, who lived at Hewlett -Packard at the time. Again, this dates me, but there was no such thing as an unlisted telephone number then, so I can just look at... I think this is really cool.
It tells the story of him cold -calling the founder of HP, asking him for some spare parts. And so maybe we can start... We can find our first...
And I was part of this group at Hewlett -Packard. When I was 12, I called up... When I was 12...
It's a great starting point for the video. When I was 12, I called up... When I was 12, I called up Bill Hewlett, who lived at Hewlett Packard at the time.
Again, this dates me, but there wasn't such thing as an unlisted telephone number then, so I could just look at the book and look his name up. And he answered the phone, and I said, Hi, my name's Steve Jobs. You don't know me, but I'm 12 years old, and I'm building up a frequency counter, and I'd like some spare parts.
And so he talked to me for about 20 minutes. I'll never forget it as long as I live. And he gave me the parts, but he also gave me a job working at Hewlett Packard that summer, and I was 12 years old.
And that really made a remarkable influence on me. Hewlett Packard was really the only... company I'd ever seen in my life at that age.
And it formed my view of what a company was and how well they treated their employees. At that time, they didn't know about cholesterol back then. But at that time, they used to bring a big cart full of donuts and coffee out at 10 o 'clock every morning.
Everyone would take a coffee and donut break. And little things like that. It was clear that the company recognized its true value as its employees.
So anyway, things led to things with Hewlett -Packard. And I started going up to their Palo Alto research labs every Tuesday night with a small group of people to meet some of the researchers and stuff. And I saw the first desktop computer ever made, which was the Hewlett -Packard 9100.
It was about as big as a suitcase, but it actually had a small capital grade 2 display in it. And it was completely self -contained. There was no wire going off behind the curtain somewhere.
And I fell in love with it. And you could program it in BASIC and APL. And I would just, for hours, you know, try to get there right up to Hewlett Packard and just hang around that machine and write programs for it.
And so that was the early days. And I met Steve Wozniak around that time, too. This is an example of a short front clip, obviously.
But just to exemplify what we're doing is we found a... worthy start to the clip. It's an interesting topic.
It's how Steve Jobs called the co -founder of HP to ask for some spare parts and got a summer internship at HP at age 12. And so this is just a great example of how you would just sit through a long form source and look for clips. And as you can see, I cut the long form source when it started and I cut at around its end.
And then what I would do to identify the clip is I would just come here and I have a key bind on my laptop that changes the color of the video file in the timeline. And so I'll just press S here and it'll make it blue. so when i go back if i want to go back through the the interview later i know that i have a clip here and what i could also do if i want to be even more specific is i could add a marker here make it a bit bigger and then name this oh steve got a job at hp at age 12.
and maybe change its color there we go so i would literally sit through this whole thing so we have three hours worth of conversations here. You would sit through the whole thing and look for clips manually if you were to do clip identification without the help of an agent, which still might be what your team is doing or might be what you were doing if you're the editor, if you're the person watching this video.
And I think there's still value in this if you're looking for some specific moment in the conversation or you want to create a specific clip around a certain topic. Agents can help you streamline this process, but I think that clip identification is still something that's slightly complicated for agents to fully grasp. And so if you want to keep doing this.
This is one of the moments where it's worth having a human in the loop, or at least reviewing the clips that the agent does. You definitely want to have a human in the loop for that. So perfect.
As we were saying, quality check and review. This is where the AI and the human get involved in the loop. So for each candidate, read the first and last 30 seconds of the transcript to catch bad boundaries.
Basically what this means is if you were to manually be using the system, not letting the system run on its own, is you would go through the clips and literally look to see if the beginning of the clip, if the start of the clip makes sense, and if it ends at a decent, like at also point. where it makes sense. It's not cutting the story short or it's not making the story longer than it needs to be.
It's just, again, reviewing the quality of the clips and the clip selection. With the system, this would be done automatically. This is trained on, again, a script that has specific guidelines and knows what to look for in a clip to make sure it's starting at the right time, ending at the right time, and it's cohesive throughout.
So point number seven, extraction, which basically, if you're in Premiere, if you're doing this by hand, just means opening your other sequence, the clip sequence, and then coming here, copying this. and then pasting this here. And now, boom, you have your first clip in an entirely new sequence.
And so if you wanted to export this, you could just command M and save all the settings and everything and just export and you'd be done. Or we can also make some adjustments to the clip first. For example, what we could do is we could have this transcribed and then we could remove the pauses and the blank spaces and things like that.
quote unquote manually with Premiere, because this is obviously a Premiere tool that doesn't require us to go through every piece of blank space and cut it manually, but doing it with the help of Premiere while having a human involved in the loop is how you would do it if you were to sit through this whole thing manually.
With the system, what it does is it uses FFmpeg, which is this, I don't know what it is actually, but it's just a tool that the computer is able to use to manipulate videos. And so FFmpeg, if you... have tried doing any AI video related stuff, you've probably run into it.
And so this is just what the system's using. to chop up the videos so extract them even cut the make the pacing of them better cut the blank spaces and all this stuff this is just something that has to exist the use clips log is just something that has to exist for the agent for the system to run properly but basically what this means again if it was a human doing it it would never happen the human has better memory when it comes to things like this but even if you were to coordinate multiple humans then this already becomes a viable processor or an interesting process it's basically just saying look i cut the clip i'm making a different color here because this clip is already used i cut the clip from 348 to 535 which is him talking about hp you store that somewhere so that the different editor when he's going through the same source material if he's going through the same source material he won't use the same clip so you're not just creating duplicates of content of clips that have already been posted in the past so i'm just going to show you now that this finished trans this finished transcribing it in premiere if we were to cut the blank spaces we would just click the filter button here look for pauses and then we can click here to define the minimum pause length but i think this one is decent and
Steve is obviously a good communicator, so he doesn't have many blank spaces and pauses when he speaks. And so we can just delete and we'll make the video a bit shorter. We can filter for filler words, which is just like, um, and things like that.
And then delete those two. And so it just makes the video a bit more well -paced, more smooth. That's how we would do it in Premiere.
So step number nine here, import plus sequences is something that Cloud has identified as eliminated or not used because the system fully, truly, fully eliminates this whole process. You don't need to do this manually anymore. But basically what it's saying is, again, by hand, creating a new premier project per batch, one sequence per raw clip, transcribing everything into text panel, yada, yada, yada, is what we're doing here.
Pacing is what I just walked you guys through with this whole explanation and this whole practical walking through the whole thing. The assembly is basically what we want to do. Say this was a clip for a business motivation mindset channel.
And in the end, we wanted to promote our ebook or our course or our cloud system for like... thinking through business, whatever, what we might do is we might come here and stitch up a CTA at the very end of the video. Or if it was a video that's like seven, eight minutes long, we can also chop up a CTA, a third into the video.
So like 30 % into the video. It also helps because you'll probably retain a lot more people by 30 % than you will by a hundred percent of the video. So you want to make sure that as many people as possible are seeing your CTA, because if they're interested, they can then take action.
So that's that. If you had a CTA, let me see if I have a CTA here, by the way. So I can just show you an example, stitch this.
This is one from a client, but you just stitch the CTA at the end of the video here. And then maybe use some sort of transition to smooth out the end of the clip with the beginning of the CTA.
There you go. And so make that. scale bigger shout out mr daniel fazio called a male wizard making an appearance in the video uh but that's that's what this step means what this is also referencing here is that if you're adding a clip 30 into the video or like a third into the video what you ideally want to do is you want to place it during a cliffhanger moment so if you say oh so this is the whole secret to create and then before you said before the person says it you just add the 10 second or the 15 second CTA.
And then you go back into the video, sort of like every commercial break on television ever does it, or it's like you wait until it's a moment that people want to consume. And then you add the CTA there because you know, they're paying attention. Now, as you can see here, step 13, what we would do next is we will get the transcript for each one of the clips.
So in this example here, I'll just go back. We already have this one transcribed, but what we would do here in premier is go back to the text panel and then export the transcript to text file. So I'm going to save this here as Steve's clip.
And what this helps us with is when we later want to create the SEO optimized captions or descriptions for the videos, we want to automatically do that with AI. And for that, we need the transcript of the video because we obviously don't want to have to listen through the whole thing and then rewrite. the description or the captions by hand.
The way the system automates this is since it already has transcribed the video before in the process, it is able to just identify what the clip is and just extract the transcript from that. specific segment and use that automatically you don't have to mess with it you don't have to find the file you don't have to put it anywhere you just identify what the clip is and it will do it automatically it will cut out that segment of the transcript automatically and work with that and so here what it's also doing is it's referencing the skills that we're using so for example optimize youtube seo is the one that creates the optimized youtube descriptions titles tags hashtags things like that to improve your discoverability and your rankings in search so for the export part of it obviously from here this is super simple you just make sure to select everything and then since we don't only have this clip in the timeline, we can just use entire source or you could do source in and out.
If you just wanted to export the clip that was selected, which is what I tend to do. You select every single one of the proper, you select all the video settings and the best export quality for your certain source. And then you just look for where you want to export it and then export it to a folder.
And what's really important here is you want to make sure that this folder that you're exporting it to is the same one you always do, because if you want to reference it, if you want to link it to cloud or to codex, you don't want to be using different folders at different times because it then has to continuously or like has to always be looking for the new folder or where's the new file where's the new video file so you want to keep things centralized as much as you can whether it's in your own local machine or if it's in the cloud you just want to have things properly organized the same principles that apply for editing in premiere will apply for everything when it comes to creating a very streamlined process with ai and for the naming convention i also incentivize having some sort of naming scheme here because it just really helps ai make things faster or identify things faster so if you have a proper if you're doing it manually and you're just posting it for yourself you obviously don't need that in the future if you want to retrieve something it might be easy
if you have a better naming scheme or naming convention. But if you don't want to do it, it's fine. If you're using AI, if you want to include AI in the loop, you obviously, you definitely want to have a good naming convention and naming scheme because it, again, it then helps the agents navigate and find things much faster and much more efficiently.
So you're preserving a lot of tokens. I just realized that I skipped. Step 14, which is related to what I was just saying, which is the numbering, like the naming scheme and everything, giving it a publish date, adding it to Notion cards.
So I use Notion to centralize this whole thing. As I was saying, you can store it in your own machine or you can store it in the cloud. I like to use Notion because it's an easy way for me to visualize everything when it's going out, who's assigned to each test or each piece of content.
Notion also helps us store all the captions, the descriptions and everything that the video needs to use. And it's also an easy place for an agent to retrieve information from. So an example of how we would integrate Notion in this process is say we have our content calendar here.
And so I'm going to create this new page, which is called Steve's clip.
And I don't have to fill this, this whole stuff out, but I'm going to copy the link to this page here. And then what I'm going to do is I'm going to go to Claude and I'm going to say, let's run optimize YouTube SEO, which is the skill on this page right here, which is the page for Steve's clip. And then what I'm also going to do is I'm going to look for the TXT file.
that i had so i have steve's clip the txt file the transcript i'm going to just put this inside notion so it knows where to look for the transcript of the video and mention that the transcript the transcript lives inside the notion page and so what it's going to do as we let it run is it's going to activate the skill start using it it's going to fetch notion read everything that we have around YouTube SEO.
Just pull in a file.
Wasn't able to actually download the text from Notion, so I think the correct way to do it would be, the best way to do it would be to have the... transcript pasted here. Since I have things running locally in my own machine, it went around, it went ahead and looked for it and downloads and found it.
So it's going to run it with the TXT file now. And so while it's at it, what we can do is we can go to Photoshop and start working on a thumbnail for our video. Thumbnails are honestly optional.
You don't have to have thumbnails on medium from clips. Great example of this again is for Mosey. He was running his channel without using any thumbnails for a long time.
Now he started doing very simple thumbnails, but I'm going to show you a very nice way to do it manually and then how to later automate it using AI. streamlining your manual stuff with AI. So I have Photoshop open here.
I created a new file that is 1280 by 720 pixels, which is the ideal size for thumbnails. And the simplest way to go about this is let's just go back through the video and find a moment where he's smiling.
Or he's making a cool face.
I think this one's cool here. So we're just going to use this camera icon here in Premiere, the export frame one.
We're going to save this as Steve's face. Obviously, I'm just saving this all in downloads. You want to have it, again, everything organized just for the sake of this example.
I'm not doing everything like I would if this was like a client project or if this was my own channel's project. I just want to be a bit faster with my demonstration. But let's save this.
And then let's go back to Photoshop. Let's look for Steve's face. And by the way, this software I'm using here is called Raycast.
So if you're on Mac, I highly recommend using it. because it really helps you navigate your files and folders and apps much faster in a much more efficient way. And so let's make this a bit bigger here.
His face is obviously very, the quality is very low. It's not good. So what we can do in Photoshop is we can go to image and then click generative upscale.
And it's going to open this little window here with Photoshop's AI. We're going to use Topaz Gigapixel, face recovery, scale 4x, and then click here to upscale this image. And let's see what it will come up with.
In the meantime, if we go back to Claude here, we'll see that it's created the whole thing using the optimized YouTube SEO skill. So it has titles for the video. Steve Jobs code called Bill Hill at 12 at 12.
The phone call they got Steve Jobs job APH 12. It's a great YouTube title. It has the full description of the video optimized for SEO.
Obviously in the case of this video, it's rare to find someone searching this up, searching this story up on YouTube, but maybe. It gives us tags, it gives us hashtags, and then it's doing an SEO check here. And now it's asking if we want to write it to Notion.
We're going to say yes. And in a second it should write the whole thing let me just delete this here it should write the whole thing to notion here and i'll show you guys what it does let's go back to photoshop and there we go it created looks a bit weird but i guess this is okay maybe we can create a different version that is just 2x upscale so it's not trying so hard to restore his face and maybe just gives it gives us a more natural result let's look at what it does Let's wait for a second.
And then let's go back to Notion in the meantime. There we go. We have everything here.
And so later, if we have an agent scheduling everything via API, they can retrieve all the needed information for scheduling the video on YouTube in Notion, which just makes things very easy for it. Yeah, I guess they both did a pretty okay job at it. Let's just use the bigger one then.
Let's go back to the main video or the main source. Because we want to keep the thumbnails size 1280 by 720 pixels. Okay, what we can also do here now that I think will be really cool is let's just make this, oops, misclicked.
Let's just make this a smart object.
Delete, maybe not the whole thing. Delete this. the black edges here and let's make this a bit bigger so it fills the screen and then what we can do is we can select these empty parts here and just generate something with ai so that it gives us it fills up the whole screen maybe this won't be necessary but i just want to see what it does there we go i think this is this is nice actually yeah this is cool let's make this into no file let's just add some guides to our design so i'm going to have steve's face in this quadrant here what we could also do is we could flip this but his face looks a bit weird when it's flipped i think this this is better And then one example of how we could do this is we could just do a quote or something.
I quote called the founder of Hewlett -Packard. Something like that. Align it to the left.
Make the text a bit bigger.
color white this is going to start looking a lot better in a second uh what we're going to do here is we're actually going to add some gradient to the side so that the text stands out a bit more oops there we go now let's put the text above this now we can see that it's a lot more just use this okay his face is still cool let's all right now let's select a subject here i'm just walking you through how you'd create a normal thumbnail for youtube video for a cut or for a for a clip so maybe we can make this we can pull this a bit higher let's select just select the subject again and duplicate this now we can add a little bit of a background, a bit darker.
Nice. I like this. Let's make the text stand out a bit more.
So let's add a bit of a drop shadow to it and then give it a bit of a, yeah, this is nice.
So you can see the text now stands out a bit more. You can put Steve on top of the gradients, but what we want to do maybe here is do a two 50 soft round.
There we go. And then put the text behind his face like this. Yes, this is cool.
So it's merging in with the image. And last but not least, what we can do is we can just give them a little bit of style so we can use camera raw. to increase the exposure a little bit, make his face a little bit brighter, increase the contrast.
That'll make it stand out a bit more. The whites are a bit too exacerbated here, so let's decrease them. Let's maybe make the shadows a bit darker.
His face is a bit orange right now, so maybe we can... Yeah, there we go.
Balance out the white levels here.
yeah this is too green this is this gives it a bit of color i like it tune up the vibrance a little bit so his face gets a bit more colorful and then we just we want to mess a little bit of the saturation but not too much because his head is already his hair is already a bit too red we can add a bit of texture to give some detail to his face clarity a little bit add a little bit of grain maybe to make it stand out a bit more this might be too much this is good uh we can obviously sharpen it but we don't need to i think this 15 is okay and then reduce the natural noise while we keep our artificial noise on top of it so there we go this is a bit more natural maybe this is too white actually But anyhow, this is a decent thumbnail for a video like this.
I call it the founder of Hewlett Packard. So now we have a very simple thumbnail made for this video. And what we could do, now this is what's cool.
Let's export this thumbnail as a JPEG because we want it to be as small as it can be. YouTube allows up until like two megabytes of size for the thumbnail. You can preserve as much quality here and it will still be fine.
Let's do this. Let's save it in downloads as Steve's thumbnail. There we go.
Now what we can do, which is really cool, is we can just go to chat GPT. Let's say we wanted to pull a thumbnail. Let's go back to Premiere from our different video, from our different source clip, which is, I just want myself to the side of it here.
The Warren Buffett one. This will be cool.
let's make him a bit bigger this is decent warren's face then we go back to chat gpt and then let's give it steve's thumbnail as a source and then warren's face and say I want you to create a new thumbnail using the Steve Jobs one as the source reference for the Warren Buffett video.
Source reference. Well, let's change the text to... and i think the video that i posted on warren buffett it was about small businesses having the edge yeah uh two let's say small businesses have the edge i'm sorry and then as specific as we can be with the thumbnail here the better the result will be we want the end product to be a 720 by 720 pixel jpeg image and i just have the jpeg image we want to upscale the quality of face dragon to the left third of the screen so the text would sit in the mid to right at the center to right third of the screen let's also add the gradient behind the text and warn so the text so it stands out a bit more and we can make the contents of the background behind him a bit darker too so that both him and the text stand out a bit more uh this is i mean i'm just riffing this off the cuff but i think this is decent so let's see what it produces and we want to use currently this is september 18th 2026 chat gpt is the best image model that we have access to and it's just easier for everyone to use because a lot of people have chat gpt and so let's just see what it does if we were to integrate this to the system as you can see in the page here when you see the version of this one that is live it will already be
updated. This wasn't properly updated yet, but the system for creating thumbnails is also integrated to the whole system. So it uses the contents of the video and the suggested titles by the optimized YouTube SEO skill to create a thumbnail factory that just spits out new thumbnails for the videos.
But this is something that, as you can see, amazing. This just, yeah, ChessBG is just insane, but it created the equivalent of this in a single prompt.
So what more can you ask? This is really, really cool. And so there we go.
This is exactly what the system would be doing, but in a completely automated way. Now we already have this. So what we want to do now, now, since we already have the thumbnail ready or the thumbnails ready, what we want to do is we also want to include them in Notion here.
so that the agent knows so the agent knows where to pull uh the thumbnail from but if we were to schedule things manually youtube this is obviously what we would do we would go to youtube studio and just tap create upload video then we'd find oops i haven't exported the video yet let me do that downloads see this clip now we're going to upload it here to youtube youtube let me just go back to notion and poppy The info phone call that got Steve Jobs a job at HP at 12.
Amazing title. We're just going to copy the description here. Paste it.
Oh, I said it was important to verify the phone number. And I didn't with this dummy account. But again, so to add custom thumbnails, you have to verify your phone number.
So just do that. And then we would add the thumbnail here. We can add it to a playlist.
Ideally, you want to add it to a playlist of a certain set of video that we're creating. It's not made for kids. this case it doesn't have paper motion uh no ai was used to improve the video in any way or to alter the video in any way uh the process of creating the video can have ai and it won't be considered ai use on youtube let's go back to notion and pull all the tabs the tags sorry language of the video we want to make it english video elements we can add subtitles we can just also add the import the end screen so ideally what i do is the best video for viewer and a button subscribe to your channel checks it's just checking everything uh to see if there's any copyrighted content in your stuff and seems to be fine or oh sorry it has found something but it has no impact to the channel uh no impact to the video okay it's fine uh and then visibility We can make it public and publish it, or we can just schedule it for, let's say, in a few minutes from now.
And there you go. The clip goes live in 25 minutes -ish. So this is how you would do it.
Again, I did it in a very freestyle -y manner. I was just pulling stuff to exemplify or demonstrate how to use the tools itself and not necessarily create. the proper clip, but I've been here for a little over an hour now.
So as you can see, this is how long the process would take if you were to do it manually, clip by clip. And obviously, the more source files you have to go through, the longer it takes, the more you have to sit through the videos, the longer it takes. And what's really cool is that we were able to do the same thing, virtually the exact same thing, just by prompting Codex and Collad to do it once.
And in a snap of a finger, it created three new videos that were automatically published to... the channel so if we go back the mapping here we have the upload schedule page and basically what this is what this is indicating here why it is important to have a human loop here is because the api currently does not support abc title tests which is something that you want to be doing for your clips it also does not support thumbnail tests and there's there was this bug where the api was heading videos to private uh by default but this is not happening anymore so this is not a problem or something to worry about and then yeah the reporting which can also be done in an automatic fashion or in an automated way is just pulling the analytics from this channel and sending it as a report in slack or whatever it's easier for you so you know what the results of your channel are so if you sat through this whole thing you probably now understand how the full process works in a lot more detail and how to actually implement this go about implementing
implementing this yourself and so as i said in the beginning of the video if you want access to this whole system all the workflows all dmd files all the prompts all the agents all the skills what you can do is you can click the first link in the description and that will lead you to a page where you can book a call with me and so we can figure out if it's better for you to implement this the diy version of this in your business and just implement it yourself and have it run if you want my help implementing it into your business and helping train your team on how to properly use it and best leverage it we can have the dwy version of this or if you just want my team running this entire thing for you end to end so you don't ever have to lift a finger.
All you have to do is give us access once, feed us the source material, and you're done, which is, by the way, exactly what we've been doing for Daniel Fazio for the past six months now. As you can see on the screen, we started posting on his highlights channel a little over six months ago, and the numbers have been growing steadily ever since, which is helping him nurture an audience of buyers, book a ton more calls, and generate a lot more revenue with content that he's already created.
Again, he doesn't ever have to lift a finger ever. We handle everything from start to finish from him. He just has a consistent stream of new videos going out frequently that is pulling from the work that he's already done.
content that he's already recorded and so yeah if you want to talk more about it once again just click the first link in the description we'll figure something out for you what works best and if not i just hope this video was valuable and you're able to extract something from it If so, leave a comment letting me know what you thought of this video.
If you think something could be better explained or if you think some part of the process wasn't as clear as others, let me know and I can potentially record a V2 of this video in the future. I just wanted to get this out as soon as possible because I was super excited about what I built and I wanted to share it with everyone.
But I definitely do plan on making an updated version of this video in the future, maybe higher production setup or something. But yeah, just excited to share. Once again, if this video was valuable, make sure to subscribe, share it with someone or leave a like and let's all create more content.
Peace.
The Hook
The bait, then the rug-pull.
The title promises a copy of a $3M channel and the first line delivers a haircut joke, then the real claim: a clips channel that runs end to end with no human in the loop, limited only by how many tokens you want to spend in a day. What follows is a rare thing in the automation genre, the same job performed twice, once by prompt in 35 minutes and once by hand in just over an hour, with the receipts for both on screen.
Named ideas worth stealing.
06:34model
The 19-stage highlights pipeline
- Channel onboarding and verification
- Source intake
- Download
- Transcription
- Declare the content type
- Clip identification
- Quality check and review
- Extraction
- Pacing cleanup
- Used-clips log
- Assembly and CTA
- Clip transcript export
- SEO titles, description, tags
- Naming and publish date
- Notion card
- Thumbnail
- Upload and schedule
- A/B title and thumbnail testing
- Analytics reporting
Each stage is written twice, by hand and with the system, with a time estimate for both. Eleven of the 19 run with no human, six still want human approval.
Steal forany repeatable production process you intend to hand to an agent or a new hire
04:10concept
You can't automate what you can't articulate
Automation is downstream of documentation. Every undocumented judgement call becomes a failure point, which is why years of manual repetition preceded the build.
Steal fordeciding whether a workflow is ready to automate at all
13:20concept
Content type sets clip length
- Coaching exchange: 3 to 10 minutes
- Podcast through line: 8 to 25 minutes
- Q&A answer: short
- Event panel: variable
A human editor reads format intuitively. An agent has to be told, because the target clip length and the shape of a good boundary both depend on it.
Steal forany prompt where the model has to guess at an unstated constraint
15:00concept
Single-pass transcript read
The agent reads the full SRT against a criteria file in one pass rather than chunking it across a swarm. Chunking saves tokens but destroys the context that makes a clip worth cutting.
Steal forany task where quality depends on seeing the whole document at once
21:00concept
The 30-second boundary check
Review only the opening and closing 30 seconds of each candidate clip. That is where almost all bad cuts live, so it catches the errors without rewatching the source.
Steal forany review step you want to make cheap enough that it actually gets done
24:40concept
The one-third cliffhanger CTA
Place the call to action around 30% into the clip instead of at the end, and drop it immediately before a payoff line so viewers sit through it the way they sit through a commercial break.
Steal forany mid-roll pitch inside long content
05:17list
DIY, DWY, DFY delivery ladder
- DIY: buy the workflows and run them yourself
- DWY: implementation plus team training
- DFY: the agency runs the channel end to end
One system priced three ways against how much of the work the buyer wants to keep.
Steal forpackaging a process you already run as a product
How they asked for the click.
VERBAL ASK
47:43product
“If you want access to this whole system, all the workflows, all the MD files, all the prompts, all the agents, all the skills, click the first link in the description and that will lead you to a page where you can book a call with me.”
Disclosed at 5:17 before any teaching happens, then repeated at the close with a six-month client channel shown as proof. Honest framing, but the pitch sits on both ends of a 50-minute video and the only path to the system is a sales call rather than a price.
MENTIONED ON CAMERA
Visual structure at a glance.
