YouTube2Text

How to Make Viral Animated Shorts Using Claude (Full Workflow) — Transcript

by Mr omi · 2,284 words · 323 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Look at these two videos. This one has
  2. 0:0229 million views and it's not a one-off.
  3. 0:05That channel is brand new and its last
  4. 0:0780 shorts got over 200 million views.
  5. 0:10The other one, that's mine. And no, it
  6. 0:12didn't come out of an AI video
  7. 0:14generator. Claude built it in 3D with
  8. 0:17code every single frame. In this video,
  9. 0:19I'll show you exactly how I made this
  10. 0:21animated short, how I trained Claude on
  11. 0:23this style frame by frame. What went
  12. 0:26wrong and how I fixed it, and how you
  13. 0:28can get it, too. You're going to get all
  14. 0:30of that for completely free. All right,
  15. 0:33let's get into it. If you've never seen
  16. 0:35this type of short, it's this. The
  17. 0:37characters are 2D, but the world around
  18. 0:39them is 3D. So, how do you train Claude
  19. 0:41to make this? First, you need
  20. 0:43references. A lot of them. I went to the
  21. 0:46channel and had Claude download the last
  22. 0:4720 shorts in this style, the ones that
  23. 0:49look this good. Now, here's the thing.
  24. 0:51If you just tell Claude to analyze a
  25. 0:53video, it won't look at every frame. It
  26. 0:55grabs a few here and there, but a normal
  27. 0:57video runs at 30 frames per second. And
  28. 0:59you've probably seen this style all over
  29. 1:01your feed. Most of those look trash. The
  30. 1:04reason is the tiny stuff. Micro
  31. 1:06animations, a little camera move, a
  32. 1:08little shading, a little light. That's
  33. 1:10what makes it feel real. And you only
  34. 1:12catch that at 30 frames per second.
  35. 1:14Every single frame, every camera move,
  36. 1:16every little change in the light. So
  37. 1:18that's the main thing I told Claude.
  38. 1:19Watch all of them at 30 frames per
  39. 1:21second. and check everything. The
  40. 1:23lighting, the contrast, the shadows, the
  41. 1:26movement, the effects, the cuts,
  42. 1:28everything, 20 shorts, about 18 minutes
  43. 1:30of video, over 32,000 frames, and it
  44. 1:33came back with a rule book. So, what's
  45. 1:35the actual workflow behind these shorts?
  46. 1:38It comes down to three layers. How the
  47. 1:40script is engineered, how the visuals
  48. 1:42are constructed, and how Claude measured
  49. 1:44all of it. First, the script. To decode
  50. 1:46it, Claude needed the audio. And the
  51. 1:48first time YouTube handed it the German
  52. 1:50dub classic. So it pulled the original
  53. 1:53English track and transcribed all 20
  54. 1:55itself. And every single one opens like
  55. 1:58this. They buried a man alive with a
  56. 2:00bell tied to his finger. He bet his life
  57. 2:02savings on the slowest horse in the
  58. 2:04race. A knight paid a priest to forgive
  59. 2:06a crime he hadn't committed yet. There's
  60. 2:08a button in the middle of town and
  61. 2:10nobody's allowed to press it.
  62. 2:11>> A paradox in the very first sentence. No
  63. 2:14intro, no setup. Then it escalates. Then
  64. 2:17comes the pivot where the real reason
  65. 2:19comes out. Then the payoff. One last
  66. 2:21line and the twist lands. And there's a
  67. 2:23signature move. 14 of the 20 flip into
  68. 2:26the explanation with the exact same two
  69. 2:28words. You see? You see? You see? You
  70. 2:31see? You see?
  71. 2:32>> And the delivery is relentless. Almost
  72. 2:34four words a second with basically zero
  73. 2:36dead air. Second, the visuals. The
  74. 2:39characters are flat, two-dimensional,
  75. 2:40but they're lit like they physically
  76. 2:42exist in that world. They catch the key
  77. 2:44light and they cast real shadows. The
  78. 2:46world is real 3D geometry traced with
  79. 2:49black ink outlines like a comic book.
  80. 2:51The colors are heavily saturated, warm
  81. 2:53highlights, cool blue shadows, and the
  82. 2:55lighting follows the story. Warm and
  83. 2:57bright while it sets things up, then a
  84. 2:59deep night blue right as the explanation
  85. 3:01begins. And the camera almost never
  86. 3:03cuts. 12 of the 20 have one hard cut or
  87. 3:06none at all. Instead, the camera
  88. 3:08physically travels into the next moment.
  89. 3:10Third, and this is the real secret, how
  90. 3:12Claude analyzed all of it. It didn't
  91. 3:14just eyeball it. It wrote its own
  92. 3:16measuring tool. Every 2 seconds of video
  93. 3:18became one contact sheet. 60 frames on a
  94. 3:21single image, 549 of them. Then it
  95. 3:24tracked the camera with phase
  96. 3:25correlation, measuring how far the image
  97. 3:27shifts frame by frame. It flagged every
  98. 3:30hard cut. And it measured the
  99. 3:31brightness, the contrast, and the
  100. 3:33saturation of every short. That's how it
  101. 3:35found that in about 80% of the frames,
  102. 3:37the camera is almost perfectly still.
  103. 3:39Just tiny drifts. So, here's the tip.
  104. 3:42Don't ask Claude what it thinks of a
  105. 3:43video. Make it measure. You can't
  106. 3:45recreate an opinion. You can recreate a
  107. 3:48number. So now I had the whole
  108. 3:49blueprint. Now the story. We made the
  109. 3:52same type of short but with our own
  110. 3:54story. Same lighting, same camera, same
  111. 3:57cuts, same script formula, just our own
  112. 3:59characters. My first topic was why time
  113. 4:02seems to slow down when you're in
  114. 4:03danger. But while Claude studied the 20,
  115. 4:06it noticed one of them already covered
  116. 4:07the exact same experiment with 5.7
  117. 4:10million views. So, we threw it out and
  118. 4:13picked a new story, one the channel had
  119. 4:15never told. 1831, 74 soldiers march
  120. 4:18across an iron bridge, all in step, and
  121. 4:20one end of the bridge drops into the
  122. 4:22river. Before writing a single word,
  123. 4:24Claude checked every fact, the date, the
  124. 4:27number of soldiers, even how deep the
  125. 4:29river was, 2 feet. And that turned into
  126. 4:31the best joke in the whole short.
  127. 4:33Luckily, the river was only 2 ft deep.
  128. 4:36Unluckily, the river was only 2 ft deep.
  129. 4:39Then it wrote the script with the exact
  130. 4:41formula from the 20. A paradox in the
  131. 4:44very first line. 74 soldiers knock
  132. 4:46[music] down an iron bridge just by
  133. 4:48walking across it. It's then it builds
  134. 4:50up. Then you see then the twist. After
  135. 4:53that, Claude started building. At first,
  136. 4:56it used Blender. Same as my very first
  137. 4:58test. Before any of this, I asked Claude
  138. 5:01for a quick 5-second test just to see
  139. 5:03the look. Claude designed its own
  140. 5:04character, a little courier. It grabbed
  141. 5:07free 3D buildings, made flat cartoon
  142. 5:09clouds, and matched the camera timing
  143. 5:11from the channel's frames. There was
  144. 5:13just one problem. On the third camera
  145. 5:15push, the camera flew right past the
  146. 5:17character. So, for the bridge, Claude
  147. 5:19started in Blender again. It tested the
  148. 5:21ink lines there and drew the whole cast
  149. 5:23in code. Soldiers, an officer, people
  150. 5:25from the town, a kid, and a parent. Then
  151. 5:28I had a hunch. Maybe these shorts are
  152. 5:30made with 3JS. That's 3D built with code
  153. 5:33right in the browser. Claude checked the
  154. 5:35frames. Flat characters standing in a
  155. 5:37real 3D world. 3JS can do exactly that.
  156. 5:40So, it moved everything over and
  157. 5:42Remotion turns it into a video frame by
  158. 5:45frame. Then, it built the world. 37 free
  159. 5:483D models. Trees, rocks, flowers, market
  160. 5:51stalls, a cart, even a windmill. And
  161. 5:54every texture, the grass, the
  162. 5:56cobblestones, the roofs, the water was
  163. 5:58painted by code. The characters got 14
  164. 6:00different faces. laughing, whistling,
  165. 6:03nervous, shocked. Then the bridge. Two
  166. 6:06towers, iron chains, and a deck 5 m
  167. 6:08wide. 74 soldiers march across it, four
  168. 6:11side by side. Every step makes it
  169. 6:13bounce, and the bounce keeps growing
  170. 6:15until it snaps. For the ink lines,
  171. 6:17Claude wrote its own effect. It finds
  172. 6:19the edges in the 3D world and draws a
  173. 6:21black line on them. About an hour after
  174. 6:23the switch, all 15 shots were working.
  175. 6:262,25 frames. But before rendering any
  176. 6:29video, Claude checked still frames. Good
  177. 6:31thing it did. In three shots, the camera
  178. 6:34was stuck inside a house. The trees were
  179. 6:36teal and a chain was right in front of
  180. 6:38everyone's faces. While Claude was
  181. 6:40fixing that, I put one of our frames
  182. 6:41next to the original, and it looked
  183. 6:43cheap. Here's what I told Claude. The
  184. 6:46original is colorful, high contrast. It
  185. 6:48looks premium. Ours looks dull. So, go
  186. 6:51analyze it frame by frame. Take your
  187. 6:53time. Use 10 agents if you have to. Just
  188. 6:56make it as close as you possibly can.
  189. 6:58And Claude agreed. Ours looked cheap.
  190. 7:00Then it broke down exactly why. The
  191. 7:03original has black ink lines on the
  192. 7:05whole 3D world. Ours had almost none.
  193. 7:07Their colors are bold with long blue
  194. 7:10shadows and deep darks. Ours were pastel
  195. 7:12and flat. Their wood and stone look hand
  196. 7:15painted. You can see the grain and the
  197. 7:17dark gaps between the planks. Ours were
  198. 7:19plain. Their backgrounds are blurry so
  199. 7:21your eye goes straight to the character.
  200. 7:23Ours were mostly sharp. And the
  201. 7:24characters, theirs have egg-shaped heads
  202. 7:26and thin little limbs. Ours had perfect
  203. 7:29circle heads and chunky bodies. They
  204. 7:31looked like clip art pasted on top. And
  205. 7:33this is where it gets interesting. I
  206. 7:35asked for 10 agents. Claude used 12. And
  207. 7:38it didn't just throw them at the
  208. 7:40problem. It built a team. Seven agents
  209. 7:42to study the gap. One for color, one for
  210. 7:44the ink lines, one for light and
  211. 7:46shadows, one for textures, one for the
  212. 7:49characters, one for depth and framing,
  213. 7:51and one for motion. The camera and the
  214. 7:53characters. Each one put the original
  215. 7:55frames next to ours and measured the
  216. 7:57difference. No guessing, real numbers.
  217. 7:59An eighth agent worked on the sound.
  218. 8:01Then four more agents to rebuild
  219. 8:03everything. One for the camera, one for
  220. 8:05the world, one for the characters, and
  221. 8:07one for the final look. But here's the
  222. 8:09problem. Four agents working on the same
  223. 8:12project can break each other's work. So
  224. 8:14Claude gave each one its own copy of the
  225. 8:16project, and each agent owned its own
  226. 8:19files. Nobody else could touch them.
  227. 8:21When they were done, Claude merged
  228. 8:22everything back file by file. So, here's
  229. 8:25the tip. When you use a lot of agents,
  230. 8:27don't just split the jobs, split the
  231. 8:29files. Then, I gave Claude 10 hours. 10
  232. 8:32hours to learn this style and make the
  233. 8:34finished video. Here's what the seven
  234. 8:36agents found. First, the ink lines. The
  235. 8:39original has black lines on almost three
  236. 8:41out of four edges. On ours, the agent
  237. 8:44found zero. The ink effect was there,
  238. 8:46but the blur effect ran in the same step
  239. 8:48and wiped it out. And the blur itself,
  240. 8:50it was so weak you couldn't see it
  241. 8:52anywhere. Next, the camera. Ours snapped
  242. 8:54from one spot to the next in three
  243. 8:56frames. The original glides over about
  244. 8:5923 frames. Slow in, slow out. Then the
  245. 9:02color. The original has no pure black
  246. 9:04anywhere. Ours went almost black. And at
  247. 9:07sunset, their shadows are purple. Ours
  248. 9:09were brown. And the characters, their
  249. 9:11bodies are 3 and 1/2 heads tall. Ours
  250. 9:13were under three. And our arms and legs
  251. 9:16were five times too thick. But the
  252. 9:17agents also found something I didn't
  253. 9:19expect. Some of our own notes were
  254. 9:21wrong. Remember the egg-shaped heads?
  255. 9:23The agent measured 38 of them. Their
  256. 9:25heads are actually rounder than ours.
  257. 9:27Ours were the egg-shaped ones. And
  258. 9:29remember the measuring tool from the
  259. 9:30start? It said the camera is almost
  260. 9:32still 80% of the time. It's not. The
  261. 9:35camera is moving in 65% of the frames,
  262. 9:38slowly pushing in. The tool just
  263. 9:40couldn't see moves that slow. And this
  264. 9:42is exactly the part other tutorials
  265. 9:44skip. I found only two other tutorials
  266. 9:46on this style. One sells it inside an
  267. 9:48app, $67 a month or $97 for everything.
  268. 9:53The other one gives you a few prompts
  269. 9:55and it says, "Your first results might
  270. 9:57not look as clean." So, here's how you
  271. 9:59can get mine. Everything from this
  272. 10:00video, every rule and every fix I save
  273. 10:03as one skill for Claude and I'm giving
  274. 10:05it away fully free on my GitHub. You get
  275. 10:08the whole repository with 5,000
  276. 10:11reference images. The link is in the
  277. 10:12description. So, you don't have to do
  278. 10:14anything. Just install it and start
  279. 10:16making videos. Everything is already set
  280. 10:18up, so the output should be as polished
  281. 10:20as it gets. And I'm going to keep
  282. 10:22uploading new things there. I'll keep
  283. 10:24updating it and making it better. And
  284. 10:26you get all of that for free. And here's
  285. 10:28one more thing. I'm making a video on it
  286. 10:30tomorrow. This realistic 3D video, as
  287. 10:33you can see, I made it. That's my name
  288. 10:34on it. Here's something wild. Pull every
  289. 10:37blood vessel out of your body and lay
  290. 10:39them in one straight line. Most people
  291. 10:41guess it would reach across a football
  292. 10:43field. Not even close. Your body hides
  293. 10:46billions of microscopic tubes branching
  294. 10:48through every muscle, every organ, even
  295. 10:50your [music] bones. Lay them all out and
  296. 10:52the line would leave your city, cross
  297. 10:54entire countries, stretch over oceans,
  298. 10:57and keep on going. [music] Scientists
  299. 10:58estimate about a 100,000 km. That's
  300. 11:01enough to wrap around the whole Earth
  301. 11:03roughly 2 and 1/2 times. Just so you
  302. 11:05know, I've never seen anything like this
  303. 11:08made with AI. Other people have made
  304. 11:09this style with Claude, but not like
  305. 11:11this. I looked and I couldn't find a
  306. 11:13single tutorial that gets this close. If
  307. 11:16this showed up on your feed, you'd never
  308. 11:18guess AI touched it. In that video, I'll
  309. 11:20show you exactly how I made it. How to
  310. 11:22animate a realistic face with an
  311. 11:24open-source AI model, how to rig the
  312. 11:26character for free on your own computer,
  313. 11:28how to make the world look real, how to
  314. 11:30cut the render from 7 and 1/2 hours to 1
  315. 11:33hour, and how to render it smaller then
  316. 11:35upscale it so it renders as fast as it
  317. 11:37gets. You can find the skill and
  318. 11:39everything in the description. The link
  319. 11:41to the GitHub repository is right there,
  320. 11:43completely free. That's all for today.
  321. 11:45Make sure to subscribe and like. And
  322. 11:47with this type of workflow, you can
  323. 11:49pretty much clone any style. Bye.

About this transcript

This page contains the full transcript of How to Make Viral Animated Shorts Using Claude (Full Workflow) by Mr omi, generated from the public captions YouTube serves with the video. The transcript has 2,284 words across 323 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.