YouTube2Text

Как Делать Вирусные Мультики-Шортсы через Claude (Полный Процесс) — Transcript

by ЗОЛОТЫЕ ЧЕКИ · 1,806 words · 272 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Take a look at these two cartoons. The
  2. 0:02one on the left got over 30 million
  3. 0:04views. And it's not just a one-off viral
  4. 0:07hit.
  5. 0:07Following that on the same channel: 21
  6. 0:09million, 16, 15, and 14 million.
  7. 0:12The one on the right is mine, and it
  8. 0:14wasn't made by a video AI, but by Claude
  9. 0:16using 3D code.
  10. 0:17Frame by frame. Today I'll show the
  11. 0:19whole process of how Claude studied this
  12. 0:21style frame by frame, how to build your
  13. 0:23own cartoon, and what went wrong.
  14. 0:25And at the end, I'll put these two
  15. 0:26cartoons side-by-side, and we'll see if
  16. 0:28it lived up to the original or if it's
  17. 0:30just a cheap copy. Right away, I’ll say
  18. 0:31I don’t have any courses. Don't message
  19. 0:33me privately about training. I’m just
  20. 0:35showing how I do it myself. Want to
  21. 0:36follow along? Subscribe to my YouTube
  22. 0:38channel. And I've posted all the prompts
  23. 0:40from this video and the ready-made skill
  24. 0:42for Claude in my Telegram channel. It's
  25. 0:44free, the link is in the description. If
  26. 0:46you haven't seen these Shorts yet, this
  27. 0:48is them: the characters are literally
  28. 0:50flat, like stickers. Round heads, stick
  29. 0:53limbs, while the world around them is
  30. 0:55real 3D.
  31. 0:56And the camera seems to fly between the
  32. 0:58objects. That combination is exactly
  33. 1:00what gets millions of views. So, how do
  34. 1:02we teach Claude to make videos like this
  35. 1:04? First, we need samples. And not just
  36. 1:07one, but many.
  37. 1:08First thing, I open code, create an
  38. 1:10empty folder for the project, give it
  39. 1:12the link to this YouTube channel, and
  40. 1:14input this task.
  41. 1:16Download the latest 20 Shorts in this
  42. 1:18style, the ones that look the best. And
  43. 1:21here's the most important part: if you
  44. 1:23just tell it to analyze the video, it
  45. 1:25won't watch every frame; it will grab a
  46. 1:27few frames from this clip and maybe from
  47. 1:29another.
  48. 1:31But a normal video has 30 frames per
  49. 1:33second, so I write to it directly: "
  50. 1:35Watch all 20 clips at 30 frames per
  51. 1:37second and check literally everything:
  52. 1:39lighting, contrast, shadows, motion,
  53. 1:41effects, and cuts."
  54. 1:43This prompt will also be available in
  55. 1:44its entirety in my Telegram. And now we
  56. 1:46have the 20 videos downloaded.
  57. 1:49And it immediately assigned 10
  58. 1:50assistants to review these very frame
  59. 1:52sheets.
  60. 1:54Why is every single frame so important
  61. 1:56to us?
  62. 1:56You've surely seen these cartoons in
  63. 1:58your feed, and most of them actually
  64. 2:00look quite cheap.
  65. 2:02It all comes down to the simplest
  66. 2:04details: a barely noticeable movement, a
  67. 2:06slight camera zoom, a touch of shadow,
  68. 2:08color, and so on.
  69. 2:10These are what make the image feel alive
  70. 2:12, and you can only catch them by looking
  71. 2:13at every single one of the 30 frames.
  72. 2:16In total, 20 Shorts, almost 19 minutes
  73. 2:18of video, over 34,000 frames, and the
  74. 2:20end result is a ready-made rulebook for
  75. 2:23Claude in the style we copied.
  76. 2:27The secret to these Shorts breaks down
  77. 2:28into three layers: how the text is
  78. 2:30written, how the visuals are assembled,
  79. 2:31and how Claude measured it all.
  80. 2:34Let's start with the most obvious text.
  81. 2:36Claude extracted the audio from all
  82. 2:38twenty videos for us and transcribed
  83. 2:39them. And more than half, 12 out of 20,
  84. 2:42start like this. A man was buried alive
  85. 2:45with a bell on his finger. He bet all
  86. 2:47his savings on the slowest horse in the
  87. 2:50race.
  88. 2:51A knight paid a priest to forgive a
  89. 2:53crime he hadn't yet committed.
  90. 2:56The paradox is literally in the very
  91. 2:58first sentence: "No intro, no greetings
  92. 3:00." And then, the tension immediately
  93. 3:02builds. Then there is a twist, where the
  94. 3:05real reason is revealed. And the final
  95. 3:07line brings the resolution. And they all
  96. 3:09have a signature technique.
  97. 3:1115 out of 20 videos transition to the
  98. 3:13explanation with the same two words,
  99. 3:15these ones.
  100. 3:16And they read this text at nearly four
  101. 3:18words per second. Meaning there are
  102. 3:20practically no pauses. The second layer
  103. 3:22is the visual itself. The scene's light
  104. 3:24falls onto the flat characters. They are
  105. 3:26tinted by the color of the place where
  106. 3:28they happen to be standing. The world
  107. 3:29casts real shadows, while the hero has a
  108. 3:31soft spot under their feet. Objects are
  109. 3:34outlined in black, like in a comic book
  110. 3:36. The colors are very saturated, with
  111. 3:39warm red-orange tones and cool, blue
  112. 3:41shadows. And most importantly, every
  113. 3:43scene has its own palette. Also, the
  114. 3:45camera barely cuts the frame. Thirteen
  115. 3:47out of twenty videos have only one cut,
  116. 3:49or none at all. The camera simply flies
  117. 3:51to the next moment itself. And the third
  118. 3:53layer, quite literally the most
  119. 3:55important. How did Claude break all this
  120. 3:57down? He didn't just eyeball it; he
  121. 3:59wrote himself a measuring tool. He
  122. 4:01turned every 2 seconds of video into a
  123. 4:03single image. 60 frames on one sheet.
  124. 4:06There are 577 such sheets.
  125. 4:08He calculated how much the image shifts
  126. 4:10from frame to frame. That is exactly how
  127. 4:12the camera itself moves. And he went
  128. 4:14ahead and marked every cut. He even
  129. 4:16measured the brightness, contrast, and
  130. 4:18saturation of every Short. According to
  131. 4:20these measurements, in 27%of the frames
  132. 4:22, the camera is literally standing in
  133. 4:24one place.
  134. 4:26Remember this number, because we'll come
  135. 4:28back to it.
  136. 4:29And here is the most important piece of
  137. 4:30advice: don't ask Claude what he thinks
  138. 4:32about a video. Just take it and make him
  139. 4:34measure it with some script. An opinion
  140. 4:37, for instance, cannot be repeated, but
  141. 4:38a number can. And that's it, our
  142. 4:40rulebook is ready. Now we make a similar
  143. 4:43Short, but with our own story. The same
  144. 4:45light, the same camera, the same cuts,
  145. 4:47and the same text formula repeated. Only
  146. 4:50the plot and the characters will be our
  147. 4:52own. My story is Clever Hans. Berlin,
  148. 4:561904. A horse that knows how to count.
  149. 4:58As long as someone who knows the answer
  150. 5:00is standing nearby. First, Claude
  151. 5:02checked the topic against all 147 shorts
  152. 5:04on the channel. They have a horse, but
  153. 5:06there are no overlaps in the other story
  154. 5:08. So, we kept the topic. Before writing
  155. 5:11the text, it verified 20 facts from the
  156. 5:13psychologist Pfungst's book, such as the
  157. 5:15date, the thirteen-member committee, and
  158. 5:17the fact that the horse was right 80%of
  159. 5:19the time without knowing the answer, and
  160. 5:2198%with it.
  161. 5:23Then it went ahead and wrote our script
  162. 5:24exactly according to the formula. A
  163. 5:26paradox in the first line. First comes
  164. 5:29the build-up, then the twist on those
  165. 5:31very two words, and finally the
  166. 5:33resolution. Next, Claude started
  167. 5:34building the video itself. And it builds
  168. 5:36it not from neural network video, but in
  169. 5:38a program that can generate 3D. This
  170. 5:40means frames assembled by code,
  171. 5:42literally right in our browser. And it
  172. 5:44can do exactly what we need. Flat
  173. 5:47characters in a real 3D world. And the
  174. 5:49Remotion program turns all of this into
  175. 5:51video, frame by frame. We couldn't
  176. 5:53download free 3D models, so Claude just
  177. 5:55drew everything with code. That is, the
  178. 5:57courtyard, houses, trees, and
  179. 5:59cobblestones.
  180. 6:00It made 10 faces for each of the eight
  181. 6:02characters, and eight separate ones for
  182. 6:04Hans the horse.
  183. 6:05From smiling and laughing to nervous and
  184. 6:08stunned. And the black outline, like in
  185. 6:10a comic book, it scripted itself; it
  186. 6:12finds the edges and traces them with a
  187. 6:13line. After half an hour, all 16 shots
  188. 6:15were working. Before rendering, Claude
  189. 6:18checked individual frames and found 17
  190. 6:20glitches where our characters went
  191. 6:22off-screen. A top hat covered a face,
  192. 6:24and the dust by the hooves looked like
  193. 6:25cotton wool. It took everything and
  194. 6:27fixed it. And this is our first version
  195. 6:29. I placed our frame next to the
  196. 6:31original. And, honestly, it looked very
  197. 6:33cheap. Then I told Claude directly. The
  198. 6:36original is bright, high-contrast, and
  199. 6:37looks expensive. But ours is dull.
  200. 6:39Analyze it frame by frame and don't rush
  201. 6:41. Use 10 agents if you have to, just
  202. 6:43make it as close as possible. And Claude
  203. 6:45agreed that ours was dull and cheap. Our
  204. 6:47shadows fall too deep into black. The
  205. 6:49camera barely moves and there is no
  206. 6:51lighting at all. And this is where the
  207. 6:52most interesting part begins. I asked
  208. 6:54for 10 agents, but Claude took six. An
  209. 6:57agent, it turns out, is a separate
  210. 6:58assistant that handles one specific task
  211. 7:00. Five of them each measure the
  212. 7:02difference in their area and redo their
  213. 7:04part themselves. That is, the camera,
  214. 7:06the world and lighting, materials,
  215. 7:08characters, and post-production. Each
  216. 7:10one puts the original frames next to
  217. 7:12ours and measures the difference. So it
  218. 7:14doesn't do it by eye, but literally with
  219. 7:16numbers. And the sixth one double-checks
  220. 7:19our own rulebook. But five agents in one
  221. 7:21project could break each other's work.
  222. 7:24Therefore, Claude gave each one their
  223. 7:25own copy of the project and their own
  224. 7:27files that no one else touches, and then
  225. 7:28assembled everything back together at
  226. 7:30the end.
  227. 7:31So, a tip for the future: when we have
  228. 7:33many agents, we shouldn't divide them by
  229. 7:35tasks, but by specific files.
  230. 7:37And here is what all our agents found.
  231. 7:40Our outline was too thick, about 11
  232. 7:42pixels, while the original is 2 to 6.
  233. 7:44Our image is also blurry. The original's
  234. 7:47sharpness is 676, while ours is 312.
  235. 7:51The camera was almost static in all 5
  236. 7:53cases, and the shadows were crushed into
  237. 7:55black. The original has almost none of
  238. 7:57that. And the agents also found errors
  239. 8:00in our own notes. Remember that 27%mark
  240. 8:02where the camera is positioned? That
  241. 8:05turned out to be untrue. It’s actually
  242. 8:07at 6 to 9%of the frames. It just crawls
  243. 8:10so slowly that the first tool literally
  244. 8:12didn't notice it. The build itself took
  245. 8:15Claude half an hour. The agent team's
  246. 8:17edits took a little over an hour. Here
  247. 8:19is what we ended up with. On the left is
  248. 8:21the 31 million original, and on the
  249. 8:22right is our video. And here is our full
  250. 8:25short.
  251. 8:39I’ve compiled everything from this video
  252. 8:41, every rule and every edit, into one
  253. 8:43skill for Claude. A skill is a
  254. 8:44ready-made instruction. You just need to
  255. 8:46put it into Claude, and it immediately
  256. 8:48knows the whole process. From breaking
  257. 8:49down clips to the finished video. It
  258. 8:51will be available for free on my
  259. 8:53Telegram channel. The link is in the
  260. 8:54description. You don't actually need to
  261. 8:56set up anything here. Just install it
  262. 8:58and make cartoons like this. And the
  263. 9:00main takeaway from this is: This process
  264. 9:02works not just with one style. Download
  265. 9:0420 clips, make Claude measure them, not
  266. 9:06just evaluate them. Put your frame next
  267. 9:09to the original. And this way, you can
  268. 9:11copy almost any style. Every week I
  269. 9:13break down formats like this on the
  270. 9:15channel. Subscribe so you don't miss out
  271. 9:17. And also don't forget to hit the like
  272. 9:18button. See you later.

About this transcript

This page contains the full transcript of Как Делать Вирусные Мультики-Шортсы через Claude (Полный Процесс) by ЗОЛОТЫЕ ЧЕКИ, generated from the public captions YouTube serves with the video. The transcript has 1,806 words across 272 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.