I found a way to make these AI animations for $0 — Transcript
Full transcript
- 0:00This channel got over 8 million views this
- 0:02month making AI cartoon series like this one.
- 0:05In the next 15 minutes,
- 0:07I'm going to show you how to make animations exactly like it for $0,
- 0:11with no Claude subscriptions and nothing to pay for.
- 0:14And here is one I made that way from a single prompt.
- 0:18<i>Peek. All the answers.</i> <i>A minute early. Mo waves. Peek.</i>
- 0:22<i>You wave first. You watched him do it.</i>
- 0:24Here is the plan.
- 0:26First, I'm going to put a free model
- 0:28and Claude Opus head to head on the same cartoon.
- 0:31So you can judge the result for yourself.
- 0:33Then I will show you how to install the free setup step by step.
- 0:38Then where to get the skill that does all the heavy lifting,
- 0:41because that is actually the real secret.
- 0:44And then we will make your first cartoon together.
- 0:46If you have seen these stick-figure What If cartoons all over YouTube and wanted
- 0:51to make your own but you didn't want to have another subscription,
- 0:55this video is for you.
- 0:57You don't need to know how to code.
- 0:58If you can copy and paste a few commands,
- 1:01you can do this and I will give you everything for free at the end.
- 1:05Quick bit about me, by the way, I am Baptiste.
- 1:07I use AI to edit all my videos and last week
- 1:10I posted a video about AI animations in Claude
- 1:13and I got so many comments asking me how you could do this for free.
- 1:17So that's why I'm making this video for you today.
- 1:20Three things had to line up for this to work
- 1:22and they only have just done that today.
- 1:24There is a free AI tool called OpenCode
- 1:27that does exactly the same job as Claude Code.
- 1:30It reads your files, runs your commands and writes code right on your computer.
- 1:35Then there is a model called GLM 5.3 Flash that is free inside
- 1:40of OpenCode right now with no account and no API keys needed.
- 1:44And then there is a bunch of skills which are folders
- 1:48of instructions and scripts that hold all the know-how.
- 1:52So the model doesn't need to be a genius. It just needs to follow the recipe.
- 1:56That last one is what makes a free model good enough.
- 1:59So let's see if the free model actually holds up against Claude.
- 2:03Here is how I tested it and I gave it as fair as I could.
- 2:07Same skills, same computer and the same one-line idea typed in,
- 2:11which is what if you could see one minute ahead?
- 2:14After that, I didn't type a single thing.
- 2:16I just let each run go all the way to the end.
- 2:20So here is the one you saw at the start made by the free model
- 2:23and here is the exact same idea made by Claude Opus.
- 2:27Overall, I think the free model looks extremely good.
- 2:31It's very close to Claude and to be honest,
- 2:33if I didn't know which one was which,
- 2:35I probably wouldn't be able to tell the difference.
- 2:38Claude Opus took about 42 minutes and the free model only took 28 minutes.
- 2:42Both of them wrote the script, voiced it,
- 2:45drew a storyboard, checked their own work,
- 2:49animated every shot and mixed the sound and both handed me
- 2:52a finished video on the first and only render at the end.
- 2:56The part that actually surprised me is what happens before the animation.
- 3:00The skill makes the model look at its own storyboard and fix anything
- 3:03that doesn't read and the free model actually did really well.
- 3:07It looked at its images, decided seven of its 15 shots just weren't good enough.
- 3:12Things like a ball sitting in front of the character's face
- 3:16or a shot zoomed in so far that the arm you're meant
- 3:19to see was out of frame and it went back
- 3:22and fixed every single one of them on its own.
- 3:25There were two real differences though
- 3:27and I think you should know about both up front.
- 3:29The first one is homework.
- 3:31The skill tells the model to search on YouTube before it writes anything
- 3:34to make sure nobody has already made that exact same idea you have.
- 3:39Opus found five what-if channels that already had done the see
- 3:43one minute in the future so it changed the idea a little bit
- 3:46to what if you could preview the next minute
- 3:49and that's why the titles are actually not the same.
- 3:52The free model searched as well but it didn't find anything
- 3:55so it just kept the original idea so
- 3:58if you are using the free model do that check yourself
- 4:01before you hit enter and start making your video.
- 4:03The second one is the sound.
- 4:06The free model went back and forth about five times trying
- 4:09to get the right mix which is where most of its effort went at the end
- 4:13but it did get there and both videos came out
- 4:16with pretty much the same loudness and the same style.
- 4:19Then we have the cost.
- 4:21Opus ran on my Claude plan and this one used
- 4:24about eight percent of my five hours limit.
- 4:27The free model cost me nothing at all you can see here we have zero dollars.
- 4:31The voice and the sound effects come from ElevenLabs in both cases and
- 4:36that was around 240 credits for the Opus one
- 4:39and about 187 for the free one.
- 4:42Good thing is ElevenLabs gives you 10,000 free
- 4:46credits every single month so that is going
- 4:48to be around 40 to 50 different animations
- 4:50you can make without paying a single thing.
- 4:53So good news is the free model can do it
- 4:56so now let's get it set up on your computer.
- 4:59Here is everything you need and what each piece does.
- 5:02OpenCode is the app the AI works in.
- 5:05GLM 5.3 Flash is the free model that does the thinking.
- 5:09Then we have Remotion which is what turns the code it writes into an actual video.
- 5:15We have ElevenLabs which does the voice and the sound effects
- 5:18and underneath all of that the skill uses three free tools.
- 5:23Node.js is what Remotion will run on
- 5:26so it's the engine that actually renders your video.
- 5:30Python runs the skill's little scripts and that's going
- 5:33to be the one that will fetch the voice,
- 5:35line the shots with the words and lay out the storyboard.
- 5:38And finally we have ffmpeg which handles the audio and video files
- 5:43so that's what gets the sound to the right volume and a finished video at the end.
- 5:48So first those three tools.
- 5:50If you are on Windows you can install all of them from the terminal with winget,
- 5:55one line each and then one more line
- 5:57for the two Python packages that the skill will use.
- 6:00But don't worry about pausing this video.
- 6:03I have all the commands for you in the description
- 6:05so you don't have to try and do this right now.
- 6:08Then OpenCode itself which is just one command as well.
- 6:12Now this is the one thing that actually
- 6:14tripped me up a little bit so don't skip it.
- 6:16Make a folder for your cartoons and always start OpenCode in that folder
- 6:21because it only sees the skills that are in the folder
- 6:25where you are actually going to open it.
- 6:27So if you open it anywhere else your skill just won't be there.
- 6:31So you need to use a cd command
- 6:33which you can see here into the folder in the terminal.
- 6:37Or on Windows you can also right click the folder
- 6:40and choose open in terminal and then just type OpenCode.
- 6:44Once you are in OpenCode you just do /models and you pick GLM 5.3 Flash Free
- 6:50and you press Ctrl+T until the reasoning level says high
- 6:54which is what I used for every run in this video.
- 6:57And that's it.
- 6:59You've now got a free AI agent on your computer that can read files,
- 7:03write code and run commands.
- 7:06But on its own it has no idea how
- 7:08to make a what-if cartoon and that is the skill's job.
- 7:13A skill is just a folder.
- 7:14Inside it there is a set of instructions written for the AI plus the scripts
- 7:19and the animation code it needs and it holds
- 7:21everything I learned making these cartoons myself.
- 7:24The story formula, the pace,
- 7:26how the light works and the checks it runs on its own storyboard.
- 7:31The model reads it and will follow it basically.
- 7:34And that's why the free model can make something this close to Opus.
- 7:37My what-if skill is free on GitHub and the link is in the description.
- 7:41Download it, put the folder inside your cartoon folder
- 7:44and put the skill in the skill folder and OpenCode will find it for you.
- 7:49I also added a small thing that makes it so much easier to use.
- 7:53In OpenCode skills don't show up when you type a slash command.
- 7:58So I added a tiny command and now you just type
- 8:01/what-if plus your idea and it loads the skill for you.
- 8:05It comes with the download and the README file shows you exactly where to put it.
- 8:09The skill does need one key though and that's your ElevenLabs key for the voice.
- 8:13Make a free account,
- 8:15create an API key and paste it in a file called .env in your folder on a line
- 8:21that says ElevenLabs API key equals and then you paste your key.
- 8:26If you don't know how to do this you can just ask GLM
- 8:29create an env folder for me and here is my ElevenLabs key.
- 8:33And you're not really just limited to my skill.
- 8:36I've also put my Instagram short form skill on GitHub
- 8:39which edits all my Instagram reels and that's how I actually post them.
- 8:44Now you might be thinking why don't I just ask the free model to build me a skill
- 8:48and actually I tested that and I'll show you what happened at the end
- 8:51because that is the one job I still ask Claude to do
- 8:55and maybe you can just also do it in ChatGPT
- 8:58if you have a free ChatGPT subscription.
- 9:00But for now let's make your first cartoon.
- 9:03Open OpenCode in your folder and type
- 9:06/what-if then your idea as a what-if question.
- 9:09Mine was what if you could see one minute ahead.
- 9:12Then you press enter and you can just walk away literally.
- 9:16In that time it's following the skill from top to bottom.
- 9:19Writing the script,
- 9:20voicing it, drawing the storyboard and fixing it,
- 9:24animating every shot in code and adding the sound.
- 9:28Mine for this video took about 28 minutes and I didn't touch it a single time.
- 9:33When it was done the finished video was
- 9:35sitting in the folder under the episode name.
- 9:38<i>Mo throws. Peek. You catch it</i> <i>before he lets go. Pop quiz.</i>
- 9:42<i>Peek. All the answers.</i> <i>A minute early. Mo waves. Peek.</i>
- 9:46<i>You wave first. You watched him do it.</i>
- 9:49<i>What if you could see one minute ahead?</i> <i>One catch. You already do this.</i>
- 9:53<i>Your brain predicts the next tenth of a</i> <i>second. That's real. So, would you peek?</i>
- 9:57<i>Peek.</i>
- 9:58And that's the whole thing.
- 10:00One line in, a finished cartoon out and a model that cost you nothing.
- 10:05Now there are a few things that will save you time.
- 10:07Start a fresh session for every animation you want to do.
- 10:12Because one animation used about 70% of the free model's memory
- 10:16and halfway through it it had to summarize its own conversation to keep going.
- 10:21And I think it coped fine but a fresh
- 10:23start will give you more room to play with.
- 10:26Do the YouTube check yourself like we saw earlier.
- 10:29Search for your idea and if a channel has already made the exact
- 10:32same idea I would highly recommend you to give it your own twist.
- 10:36Check the fact.
- 10:37Every episode has one real fact in it
- 10:39and the skill makes the model write down where it came from.
- 10:43The free one cited a real paper from a journal for mine.
- 10:47But open the source and check it before you post your video.
- 10:51If something looks off just tell it in plain words.
- 10:54Something like the ball in the shot was covering his face
- 10:57and it will fix that shot then render it again.
- 11:01And the same goes for the characters.
- 11:04You probably don't want Pip, my character,
- 11:07in every single of your episodes and you don't have to keep him.
- 11:10The characters in this skill aren't actually generated images.
- 11:14They're all drawn by the same bit of code with a different costume on top.
- 11:19So making your own dress really means describing what you want.
- 11:22For example I typed add a new hero to the cast,
- 11:26a girl with a high ponytail, a yellow raincoat and red boots.
- 11:29And about six minutes later it showed me my new character in the cast sheet
- 11:34with my new hero standing next to everyone else so I could check if they fit in.
- 11:39And because the character is just code it looks exactly
- 11:42the same in every shot of every episode you make.
- 11:46And that means there is no drift.
- 11:47It's going to have that consistency you need.
- 11:50Then I told it to make her my hero for the minute ahead episode instead of Pip.
- 11:55And about six minutes later here is the same cartoon with her in it.
- 11:59It came the same voice and the same sound so that didn't cost me a single credit.
- 12:04By the way if you want to go really deep on how this look
- 12:07and these characters were designed
- 12:08in the first place I made a whole video on that and I'll point you to it at the end.
- 12:13And that brings me to the one job I still
- 12:15think will be hard for the free model to do.
- 12:18It's building the skill itself.
- 12:20To test it I gave the free model frames from one of the viral What
- 12:24If episodes and asked it to break down the style.
- 12:27Which is exactly how I started this skill with Claude.
- 12:31Three minutes later it came back with 22 rules
- 12:34and honestly it got most of the look right.
- 12:37The sunset color, the light coming from the window and the lamps,
- 12:41the glow around the edge of the characters,
- 12:45the blur, the foreground and it even noticed the episode
- 12:48is letterboxed which my skill doesn't actually do.
- 12:51But it got the pace quite wrong.
- 12:54It counted about 16 cuts a minute
- 12:56and the real number in the opening is 27 cuts per minute.
- 13:01So a skill built on its notes would make cartoons
- 13:03at about half of the speeds of the one going viral.
- 13:06And that breakdown is the easy part.
- 13:09Trying to get into a skill that actually works took me
- 13:12about an hour of building and that was fixed in Claude.
- 13:16So this is where I have landed.
- 13:18The free model is great at following a recipe
- 13:21and Claude is still what I personally use to write that skill.
- 13:25The good news is you don't have to write it because I will give you mine.
- 13:29And if you want to look for other skills on the internet you will find
- 13:33plenty of videos giving you these kind of skills completely made for you.
- 13:37You can find them on Twitter, on YouTube, on Instagram.
- 13:40So you don't need to spend too much time trying to
- 13:44recreate your own skill.
- 13:45Everything I used is linked in the description.
- 13:48The skill, the commands,
- 13:50the setup steps and tell me in the comments
- 13:52if you have made something yourself.
- 13:54I would love to see it and maybe give you some feedback as well.
- 13:57And like I said earlier, if you want
- 13:58to see how this skill was built in the first place
- 14:01with Claude from studying the channels behind these cartoons
- 14:05to design the look and the characters, watch this video next.
About this transcript
This page contains the full transcript of I found a way to make these AI animations for $0 by Baptiste ODonovan, generated from the public captions YouTube serves with the video. The transcript has 2,679 words across 258 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.