YT YouTube2Text

YouTube2Text → Transcripts → AI Animation Pipeline: How I Make Cartoons 100% With AI

AI Animation Pipeline: How I Make Cartoons 100% With AI — Transcript

by Sanji Nai-Chien · 2,600 words · 410 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I just created this full cartoon in only
0:0220 minutes with no animation skills and
0:04I only used AI. So for years making a
0:08cartoon meant mastering drawing,
0:10animation, editing, voice acting, and
0:12somehow keeping everything consistent
0:15from episode to episode. Simply put, it
0:18was too much work for just one creator.
0:20But today, that's no longer true.
0:23Studios like Netflix, Disney, and Warner
0:26Brothers are already using AI to speed
0:28up production from concept art and
0:30character design to animation [music]
0:32and rendering. But here's where things
0:34get really interesting. Many of the same
0:36capabilities that used to require
0:38millions of dollars in budget are now
0:40available to literally anybody. [music]
0:42And I'm not exaggerating. I can tell you
0:45with certainty that this is the first
0:46time in human history when one person
0:49can control a production pipeline that
0:51used to require an entire animation
0:53studio.
0:54So, in the next 20 minutes, I'm going to
0:56show you how you can from scratch write
0:58a captivating script, build characters
1:01with distinctive voices, and create a
1:03world around them that looks original.
1:05So by the end of this video, you'll have
1:07a real unique cartoon with a complete
1:09story and you'll do it all in one
1:12sitting. So without further ado, let's
1:14get into it. Now before we even generate
1:16a single image, we first need to decide
1:18what this world is actually going to
1:20look like. One mistake that I often see
1:22people making is jumping straight into
1:24the prompting. But honestly, every great
1:26animated show out there starts with art
1:28direction first. [music] So before you
1:30create a character, you need to define
1:32the visual language that underpins the
1:35series. So here's what I recommend.
1:37Spend some time collecting references.
1:39Now Pinterest is great for this, but you
1:41can use Instagram, ArtStation, [music]
1:43basically anywhere that you can find
1:45artwork that resonates with you. Don't
1:47look for one perfect image, instead
1:50build a library of styles, colors,
1:52character designs, and environments that
1:54all fit the kind of world that you're
1:56looking to create. But, to save you a
1:58few hours, I've already put together a
2:00huge collection of references that you
2:02can use as a starting point. You can
2:04find the entire folder linked in the
2:06description [music] below. So, once you
2:07have your visual direction locked in,
2:09then let's start building. Now, it's
2:12time to find the character who's going
2:13to carry your entire series. And
2:15honestly, don't think of this as just
2:17generating an image, think of it as
2:19actually casting your lead actor.
2:20[music]
2:21You're going to animate this character
2:22over and over again, so don't just
2:25settle for the first result. Start by
2:27uploading two things: a few images of
2:29your character or yourself if you want
2:31to be the main character, and one of the
2:33style references that we picked earlier.
2:35One's going to tell the AI who to
2:36create, while the other will tell it how
2:39they should look. Then, you need to
2:41describe everything that you want to
2:42stay consistent. [music]
2:43Hair, clothes, accessories, eye color,
2:46anything that you leave out will become
2:48[music] something that the AI has to
2:49interpret. Generate multiple variations,
2:52compare them side by side, and do not
2:55stop until one immediately feels like
2:57the face of your show. I've actually
2:59also included the exact prompt template
3:01to use in the description, so all you
3:03need to do is replace my character
3:04description [music] with yours. Now,
3:06once you've found your character, then
3:08we can make them consistent from every
3:10single angle. Like I said, now that we
3:12have our character, our goal is to make
3:14sure they stay consistent throughout the
3:15entire series. Now, just one great image
3:18is not enough. The AI still will only
3:21know what your character looks like from
3:22that specific angle. As soon as you ask
3:25for a different expression, different
3:26pose, or change the camera angle, then
3:29it's going to start filling in the
3:30missing details on its own. And that's
3:32exactly where you're going to start to
3:34experience drift. By now, you already
3:37know the rule. Whatever you don't
3:38define, the AI will hallucinate. A
3:41character sheet is simply where we lock
3:43in those decisions once, so every future
3:45generation can follow the same
3:47blueprint. [music] Use the image that
3:49you just created as your reference, make
3:51any final adjustments to the design, and
3:53then generate your character sheet.
3:55Honestly, this is also the perfect time
3:57to fix anything that you don't like
3:59before you move [music] on. It's much
4:00easier to do this now than after you've
4:03already generated 20 different scenes.
4:05So, from this point on, your character
4:07sheet becomes the foundation for every
4:09image and every video that you create.
4:11Instead of redefining your character
4:13every time, you're giving the AI a
4:15single source of truth that it can
4:17reference throughout the entire series.
4:20Now that we've locked in our main
4:21character, we can start expanding our
4:23world. Instead of just starting from
4:24scratch every time, use your character
4:26sheet as the reference. This is going to
4:28keep every new generation in that exact
4:31visual style while letting you create
4:33completely new assets. For example, I
4:36can generate a dog that matches the
4:37style we've already built. Later on,
4:40I'll use that same workflow to create
4:42locations, props, and honestly
4:44everything else that the story might
4:45need. And because they're all generated
4:47from the same [music] visual foundation,
4:50everything naturally feels like it
4:52belongs together. And you'll notice that
4:54as you do this, your asset library is
4:56going to grow very quickly. So, that's
4:58why I highly recommend organizing
5:00everything from the start. Every time
5:02you generate something that you're happy
5:03with, whether it's a character, a
5:05location, or a prop, make sure to save
5:07it to your favorites. And trust me,
5:09[music] this is going to save you a lot
5:12of time down the line. You'll be able to
5:14instantly find the exact asset you're
5:16looking for, and it's going to do the
5:18work for you. By the end of this step,
5:20you should have what's a small library
5:22of organized characters, environments,
5:24and props [music] that are ready to use.
5:26That preparation is what makes the rest
5:28of the production process dramatically
5:30faster. Now that we've built our
5:32characters, it's time to build the sets
5:34that they'll actually be performing. So,
5:36just like your characters, your
5:37locations need to stay consistent
5:39throughout the entire series. I mean,
5:41[music] think about shows like The
5:42Simpsons. You can instantly recognize
5:44their living room, the kitchen, even the
5:46front yard to the house. Those recurring
5:48locations are part of what makes the
5:51world-building complete. The workflow is
5:53almost identical to what we did with our
5:55characters. Start by referencing your
5:57character sheet, so the AI keeps the
5:59same visual language. Then, generate the
6:01master version for your location. Now,
6:04in my case, I'm creating a cozy living
6:06room, but the same exact approach is
6:08going to work for literally any
6:10environment that your story needs.
6:12Once you've locked in that location,
6:13don't stop there. Generate multiple
6:15views of the exact same environment.
6:17[music] Think about all the camera
6:19angles that you'll actually use when
6:20you're telling your story. Maybe one
6:22shot is going to focus on the sofa,
6:24maybe another will look towards the
6:25window, another perhaps is from above,
6:28and the last one is from the floor where
6:30your character might be sitting. Now,
6:32every single one of these is it going to
6:34become a reusable shot that you can come
6:36back to time [music] and time again.
6:39This is where your production pipeline
6:41is going to start to pay off. So,
6:42instead of generating every scene from
6:44scratch, you can now combine your
6:46character sheet with one of your
6:47locations. The AI is already going to
6:50know exactly what your character and the
6:52environment should look like. All that
6:54you're doing is telling it which
6:55character goes to which location. Now,
6:58this is going to give you a huge amount
7:00of control over the final result. Your
7:02character is going to stay consistent,
7:04your location is going to stay
7:05consistent, and every single scene feels
7:08like it's happening inside of one world
7:10instead of a completely new one every
7:12single time. Of course, if you just need
7:14a quick getaway, you don't always need
7:16to build an entire location sheet. You
7:18can simply reference your character
7:19sheet and describe the environment that
7:21you want. That's perfect for one-off
7:23shots or scenes that appear just one
7:25time. But, for locations that you're
7:27going to revisit throughout the entire
7:29series, spending a few extra minutes up
7:31front building proper location and
7:33character sheets will help save you
7:35hours later on. At this point, you've
7:38built a complete library of reusable
7:40assets, consistent characters, reusable
7:42props, and repeatable locations. By this
7:46point, every character has a consistent
7:48appearance. The final step is just
7:50giving them a consistent voice. And in
7:53the same way that people recognize a
7:54character by the way that they look,
7:56they'll also recognize them by the way
7:58that they sound. A unique voice makes
8:00every conversation feel that much more
8:02believable. For this, we'll head over to
8:04the audio tab and open Seed Audio. You
8:07can either choose one of the built-in
8:08voices or upload your own if you want
8:10your character to sound just like you.
8:12Now, what I'll usually do is generate
8:14the same line a few times. [music]
8:16And I'll do something simple like,
8:18"Where's the TV remote?" And I'll
8:19compare the results until I find the
8:21voice that fits my character best. Once
8:23you've found the right one, make sure to
8:25download it. We're going to use it every
8:27single time throughout the series, so
8:29it's worth spending those few extra
8:31minutes making sure you got it right.
8:33Now that we've created our voices, it's
8:35time to add them to our asset library.
8:37Up until now, we've been building
8:39reusable visual assets. Think about our
8:41characters, our locations, and our
8:43props. Voice references actually work
8:45[music] the exact same way. Simply
8:48upload every voice you've generated into
8:49your assets, give each one a clear name
8:52so that you can instantly tell which
8:54character belongs where, and from this
8:56point on, every major part of your
8:58production pipeline is now [music]
9:00reusable. Your characters are saved,
9:02your locations are saved, your props are
9:04saved, and now [music] your voices are,
9:06too. That means that every time you
9:08build a new scene, you're no longer
9:10starting from scratch. You're simply
9:12pulling the right assets together to
9:14tell the next part of your story. Now
9:16that we've built all of our reusable
9:18assets, it's time to finally put them to
9:20work.
9:21>> [music]
9:21>> Now, for this part, I'm using Claude
9:23with a custom skill that I built for
9:24this exact workflow. A skill is
9:26basically a set of instructions that
9:28Claude is going to follow every single
9:30time. And mine already contains our
9:32script, character sheets, location
9:34sheets, and voice references. So, Claude
9:37doesn't just know what I want, it knows
9:39the entire world that we're working in.
9:42And yes, it's linked below in the
9:43description, so you can use the same
9:45exact one. All that I need to do is
9:48write a simple description of what
9:50should [music] happen. Now, in this
9:51case, our character is lying on the
9:52couch and they're looking for a
9:54mysterious remote hidden between the
9:56cushions. Once he finally finds it,
9:58every button [music] pressed instantly
10:00transports him to a different point in
10:02time. That's it. That's the entire
10:04prompt that I fed Claude. This skill is
10:07going to take one sentence and turn it
10:09into a complete production prompt. It's
10:11going to plan the camera shots, break
10:13the scene into time-stamped moments,
10:15write the visual directions, and place
10:18every single line of dialogue exactly
10:20where it belongs. Now, by default, it
10:22structures everything as a 15-second
10:24generation.
10:25>> [music]
10:25>> If you want a shorter scene, just tell
10:27Claude make it 5 seconds and adjust
10:29everything else automatically. Once the
10:32prompt is ready, we send it over to
10:34Higgs Field, we'll attach our
10:35references, the character sheets for
10:37everyone in the scene, the location we
10:39want, and the voice references that
10:40we've prepared earlier. But here's the
10:42part most people get wrong. Don't just
10:44upload your references and hope that the
10:46AI figures out what it needs to do. You
10:49need to tell it explicitly. And that's
10:51what the at key is [music] for. Now,
10:53inside the prompt, whenever you mention
10:55a character or a location, hit the at
10:58sign and tag the exact reference you
11:00want. So, instead of writing show the
11:02man sitting on the sofa, you're writing
11:04show image one sitting on the sofa from
11:06V1 shown in image two. And that same
11:10exact rule is going to apply to your
11:12voices. After every single line of
11:14dialogue, hit the at sign and tag the
11:16voice that should be saying it.
11:18>> [music]
11:18>> Now, this feels like a tiny detail, but
11:20in reality, this is exactly what keeps a
11:23multi-character conversation from
11:25falling apart. Skip the tags and the
11:28model is going to start mixing voices
11:30halfway through the scene. Now, when
11:32everything is tagged, then we can
11:34finally generate.
11:39Another long day.
11:42>> [snorts]
11:47>> Where is it?
11:57This is not a TV remote. Whatever.
12:19>> [screaming]
12:20>> I CANNOT PRESS IT.
12:32INTRUDER, seize him.
12:38Break it.
12:46>> [screaming]
12:51[music]
13:02>> Okay. Okay. That really happened.
13:06>> [sighs and panting]
13:09>> You stay in there. Bedtime.
13:14No more TV. Every new generation now
13:17feels like another episode of the same
13:20show instead of a completely different
13:22project. One thing that I've learned
13:24after generating hundreds of scenes is
13:26that you almost never need to start over
13:28from scratch. Honestly, if a shot isn't
13:30perfect, I'll usually make a small
13:32adjustment instead. Maybe I'll swap a
13:34key visual, rewrite a single action, or
13:37I'll regenerate just one scene instead
13:39of the entire sequence. That's one of
13:41the biggest advantages of building your
13:43assets first. Because your characters,
13:45locations, voices, etc. are already
13:47locked in, every iteration is going to
13:50build on the same foundation instead of
13:52creating something completely different.
13:54Now, you'll also notice that some scenes
13:56naturally work better than others. So,
13:59when you're choosing your key visual,
14:00try to avoid anything that's
14:02unnecessarily difficult for the model to
14:04understand. Simple, readable
14:06compositions almost always [music] will
14:08produce better results. And if a
14:10generation has a small mistake at the
14:12beginning or at the end, don't be afraid
14:14to fix it in the edit. Sometimes a
14:17simple trim is all it takes to turn a
14:19good generation into a great one. And
14:22here's the biggest mindset shift.
14:23Animation studios always build their
14:26production pipeline first. [music] They
14:27create characters, environments, props,
14:30voices, and the visual language that
14:32underpins the world. [music] Then, every
14:34new episode is simply another story
14:36that's been built on top of those same
14:38assets. [music] That's actually exactly
14:40the same thing that we've built here.
14:42Once your pipeline is in place, creating
14:45a new episode doesn't mean rebuilding
14:47your entire world. It actually just
14:49means opening Claude, describing the
14:51next scene, relaxing, and letting the
14:53workflow do the rest. The skill I showed
14:56you throughout this video is actually
14:58linked below if you want to try it for
15:00yourself. So, as always, thank you guys
15:02so much for watching, and I'll see you
15:04in the next one.

About this transcript

This page contains the full transcript of AI Animation Pipeline: How I Make Cartoons 100% With AI by Sanji Nai-Chien, generated from the public captions YouTube serves with the video. The transcript has 2,600 words across 410 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.