YouTube2Text

9일 만에 210만 뷰, 3D 사건다큐 무료로 만드는 법 (클로드 코드 + 블렌더) — Transcript

by 불혹청년 · 3,862 words · 654 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This channel reached 2.1 million views
  2. 0:01and 20,000 subscribers in just 9 days
  3. 0:03with its first video.
  4. 0:05It’s a true crime and cold case
  5. 0:07documentary channel where small dolls
  6. 0:08recreate crime scenes.
  7. 0:10A few days later, they uploaded a second
  8. 0:1219-minute video.
  9. 0:13That one also surpassed 90,000 views in
  10. 0:15just three days. After seeing this, I
  11. 0:17thought, "Should I try something like
  12. 0:18that too?" I've always liked mystery
  13. 0:20cases, so I decided to start a new
  14. 0:22channel.
  15. 0:23The channel is called "Cold Case Model
  16. 0:25Room." I have currently uploaded episode
  17. 0:271 and episode 2.
  18. 0:29All 3D scenes were created using a free
  19. 0:31tool called Blender.
  20. 0:32There is no AI used in the videos at all
  21. 0:34. So, there are no additional costs
  22. 0:37other than using Claude. However, as I
  23. 0:39made it for the first time, there was
  24. 0:40one thing I found lacking.
  25. 0:42The dolls wouldn't move their limbs.
  26. 0:45The first video of this Korean channel
  27. 0:46was the same way.
  28. 0:49But looking at their second video, this
  29. 0:51time a person is walking around. They
  30. 0:53had fixed the same thing I noticed. I
  31. 0:55had the same dilemma. So I did some
  32. 0:57research, applied some free motion
  33. 0:59capture data, and now they can sit and
  34. 1:01answer the phone like this.
  35. 1:04So, can you all try copying this channel
  36. 1:05right now?
  37. 1:07Today, I will explain why this channel
  38. 1:09blew up, how I made episodes 1 and 2,
  39. 1:11and even give you a kit so you can make
  40. 1:13the same thing on your own computer.
  41. 1:17And the fourth episode I'm currently
  42. 1:18working on actually looks much better
  43. 1:20than the Korean channel I referenced.
  44. 1:22I will reveal the secret to that as well
  45. 1:24.
  46. 1:25But this Korean channel isn't the one
  47. 1:26that invented this format. The original
  48. 1:28creator is an overseas channel called "
  49. 1:30Fern." It means "fern" in English. They
  50. 1:33have over 5.6 million subscribers. One
  51. 1:35of their signature videos has surpassed
  52. 1:3720 million views.
  53. 1:38It is a 23-minute video about the true
  54. 1:41story of teenage hackers who hacked the
  55. 1:43CIA.
  56. 1:44If you look, most of the screen is dark
  57. 1:46, and light falls like a puddle only
  58. 1:47where the doll is standing.
  59. 1:49It feels like looking down at a
  60. 1:51miniature set from above. The Korean
  61. 1:53channel brought the "Fern" style over to
  62. 1:54the Korean language. The dark stage, the
  63. 1:56faceless dolls, and even the low, calm
  64. 1:58voice.
  65. 2:00I asked Claude to analyze five Fern
  66. 2:02episodes and the Korean channel's videos
  67. 2:04in their entirety. I had it look at how
  68. 2:06many times the scene changes, how fast
  69. 2:08the speech is, and even how the sound is
  70. 2:09layered.
  71. 2:11First, the subject matter is true cold
  72. 2:13cases. "Who was the culprit?" That
  73. 2:15single question keeps viewers hooked
  74. 2:17until the end of the video.
  75. 2:18Even when it's over 10 minutes long,
  76. 2:20people don't leave. The fun part is the
  77. 2:21comments. The most-liked comments aren't
  78. 2:24even about the culprit.
  79. 2:25"Why did that person go out at dawn
  80. 2:27without their glasses?"
  81. 2:28Thousands of people start deducing
  82. 2:30theories over one small question like
  83. 2:31this. Even after the video ends, the
  84. 2:33investigative meeting continues in the
  85. 2:34comment section.
  86. 2:36And this comment also got over 1,000
  87. 2:38likes.
  88. 2:39This is their first video ever? People
  89. 2:41were truly surprised. Second, the
  90. 2:43faceless dolls. You don't have to draw
  91. 2:46or find photos of real people, so it's a
  92. 2:48huge relief.
  93. 2:49You don't have to use victims 'faces
  94. 2:51carelessly, and it even makes the
  95. 2:53visuals creepier.
  96. 2:55The third one was the most surprising.
  97. 2:57It's the sound. Pun keeps the background
  98. 2:59music much quieter than the voiceover.
  99. 3:01Almost to the point where you can't hear
  100. 3:02it.
  101. 3:03But at crucial moments, a door slamming
  102. 3:05or footsteps pop out as loudly as the
  103. 3:07voice.
  104. 3:09Since it's quiet most of the time, that
  105. 3:10one moment really hits you.
  106. 3:13My first video was the opposite.
  107. 3:15I layered background music and added
  108. 3:17tons of sound effects, but they were all
  109. 3:18the same volume, so nothing could be
  110. 3:19heard.
  111. 3:21Fourth, the thumbnail we saw earlier. It
  112. 3:24doesn't explain with text, it just asks
  113. 3:25: "What happened to this person?" It
  114. 3:27throws just one question at the viewer.
  115. 3:29And fifth, this is something I didn't
  116. 3:31know until I broke it down. I thought
  117. 3:33these videos were 3D from start to
  118. 3:35finish.
  119. 3:36But when I counted the scenes one by one
  120. 3:38, 3D made up just over half of the 117
  121. 3:40scenes.
  122. 3:42The remaining 40%or so were text cards
  123. 3:44for maps, people, relationship charts,
  124. 3:46and timelines.
  125. 3:48Their second video went a step further
  126. 3:50and even mixed in actual footage from
  127. 3:52the scene. The screen is also very dark
  128. 3:54.
  129. 3:55They show the events in 3D and use cards
  130. 3:57to clarify confusing parts.
  131. 4:00So even if it's a case you're seeing for
  132. 4:01the first time, you don't get lost.
  133. 4:04This also means you don't have to create
  134. 4:06100%3D content.
  135. 4:08I even measured their speech. About five
  136. 4:10syllables per second. It's quite slow.
  137. 4:13And there are 15 pauses that last longer
  138. 4:15than a second.
  139. 4:17They take a beat right before the scary
  140. 4:19parts.
  141. 4:20To summarize: one question, faceless
  142. 4:23dolls, quiet sounds with a sudden "thud
  143. 4:25," text-free thumbnails, and half 3D,
  144. 4:28half cards.
  145. 4:30So I decided to use these five as my
  146. 4:32framework.
  147. 4:33But I'll change the outer appearance to
  148. 4:34suit my style.
  149. 4:35However, the truly difficult part was
  150. 4:37creating the 3D visuals that followed.
  151. 4:41So, should you try copying this? Looking
  152. 4:44at the pros, the growth rate is
  153. 4:45incredible.
  154. 4:47I mean, 2.1 million in nine days.
  155. 4:49Once you build one set, you can keep
  156. 4:51filming different cases, so it's easy to
  157. 4:53upload weekly.
  158. 4:55There are only a handful of channels
  159. 4:56like this in Korea.
  160. 4:58There are many true crime channels, but
  161. 5:00most just use photos and voiceovers.
  162. 5:02Places that recreate things with 3D
  163. 5:03dolls are rare.
  164. 5:05Also, because each episode features a
  165. 5:07different case, there’s a low risk of
  166. 5:08YouTube flagging it as repetitive, "
  167. 5:10cookie-cutter" content.
  168. 5:12But there is one thing that holds me
  169. 5:13back.
  170. 5:14Honestly, it’s quite difficult to make.
  171. 5:16I mean, it's 3D. Making puppets,
  172. 5:19building stages, setting up cameras and
  173. 5:21lighting by hand takes hours for every
  174. 5:24single scene.
  175. 5:26This would be a great channel if only it
  176. 5:28were easier to make, but that’s the
  177. 5:29hardest part. So, I decided to entrust
  178. 5:32this entire difficult process to Claude
  179. 5:34.
  180. 5:36I’ve created two episodes.
  181. 5:38I’ll show you a preview here, and for
  182. 5:40those who want to see the full ending,
  183. 5:42I’ve left a link in the description.
  184. 5:45One episode is about a 1986 incident in
  185. 5:48Los Angeles where the culprit was caught
  186. 5:50after 23 years.
  187. 5:52The other is the 1968 300 million yen
  188. 5:54robbery in Japan, where a cash transport
  189. 5:56vehicle was stolen entirely.
  190. 5:58The production process goes like this.
  191. 6:00Claude writes the script, and then I add
  192. 6:02the voiceover using PC audio.
  193. 6:04Claude creates a table for every scene,
  194. 6:06detailing which stage to use, who is
  195. 6:08where, and where the camera should shoot
  196. 6:09.
  197. 6:10A free 3D program called Blender reads
  198. 6:12that table to build the stage and take
  199. 6:14the shots.
  200. 6:16Finally, a tool called Remotion adds the
  201. 6:18subtitles and text cards.
  202. 6:20From start to finish, I only had to give
  203. 6:22instructions in Korean.
  204. 6:24However, when I actually tried making it
  205. 6:25, there were more than just one or two
  206. 6:27problems.
  207. 6:28The puppets' fingernails kept flickering
  208. 6:30on screen, the visuals were flat, and
  209. 6:32the colors were almost grey.
  210. 6:34When I placed it side-by-side with Fun,
  211. 6:36the difference was immediately obvious.
  212. 6:38So, Claude and I identified the causes
  213. 6:40one by one and fixed them all.
  214. 6:44The flickering stopped, I changed the
  215. 6:46camera to orbit the puppets, which
  216. 6:47tripled the movement on screen, and
  217. 6:49adjusted the colors to match Fun.
  218. 6:52I’ve applied all these fixes and
  219. 6:54included them in the installation kit.
  220. 6:57You won’t have to go through the trial
  221. 6:59and error that took me days to resolve.
  222. 7:02If you use the kit, it comes with
  223. 7:04everything fixed from the start.
  224. 7:05You can get the kit from the Brop Youth
  225. 7:07group chat that I’m running.
  226. 7:10I plan to keep updating it as we find
  227. 7:12more issues, so please subscribe and
  228. 7:14join the chat.
  229. 7:16I also changed the audio based on what I
  230. 7:17learned from Fun earlier. I lowered the
  231. 7:19background music significantly below the
  232. 7:21voiceover and only boosted sound effects
  233. 7:23for moments like doors closing or
  234. 7:24cameras flashing.
  235. 7:26I made sure to include at least six such
  236. 7:27high-impact moments per episode.
  237. 7:29I also added a few of my own unique
  238. 7:31touches.
  239. 7:32An evidence board with red strings
  240. 7:34connecting suspects, and a map spread
  241. 7:36out on a table.
  242. 7:37And a transition where a period
  243. 7:38newspaper photo smoothly transforms into
  244. 7:40a 3D scene with the same composition.
  245. 7:42For the 300 million yen case, since the
  246. 7:44statute of limitations date is important
  247. 7:46, I added a clock counting down to that
  248. 7:48day.
  249. 7:49I put the script through a custom
  250. 7:51evaluation system I built, and only
  251. 7:53recorded it after it scored 95 out of
  252. 7:55100.
  253. 7:56Because if the script is weak, people
  254. 7:58will leave no matter how good the
  255. 7:59visuals are.
  256. 8:01After these refinements, it’s become
  257. 8:03quite similar to Fun, the original
  258. 8:04overseas channel. But there is one thing
  259. 8:06I couldn't fix until the end.
  260. 8:09Punn's puppet walks, turns its head, and
  261. 8:12opens the door. My puppet is stuck in a
  262. 8:14pose and doesn't move at all.
  263. 8:17To animate this by hand, you have to
  264. 8:19adjust elbows, knees, and heads frame by
  265. 8:21frame, which is the most time-consuming
  266. 8:23task in 3D.
  267. 8:25That's when I found Motion Bricks,
  268. 8:27released by NVIDIA this year.
  269. 8:29They announced it at the world's largest
  270. 8:31computer graphics conference and made it
  271. 8:33free for anyone to use. The principle is
  272. 8:35simple.
  273. 8:36If you provide the current pose and the
  274. 8:38final pose, the AI fills in the natural
  275. 8:41human movements in between. It’s like
  276. 8:43telling a person standing up to go sit
  277. 8:45in that chair, just giving it a
  278. 8:46destination.
  279. 8:48It creates short segments of around one
  280. 8:49second at a time.
  281. 8:51It uses the end pose of that segment as
  282. 8:53the start of the next and stitches them
  283. 8:54together.
  284. 8:55So, long movements like walking from a
  285. 8:58door to a chair come out seamlessly.
  286. 9:00The computer does in seconds what would
  287. 9:02take a person half a day to do by hand.
  288. 9:05And there is one more source of motion
  289. 9:07data.
  290. 9:09It’s a real human motion library called
  291. 9:11Boneseed, containing 70,000 real
  292. 9:13movements of people waiting, calling,
  293. 9:15and looking around—140,000 if you count
  294. 9:17the mirrored versions.
  295. 9:20When I transferred this to my puppet,
  296. 9:21this is what happened.
  297. 9:22It stands, waits, looks around, and
  298. 9:24crosses its arms.
  299. 9:25That same mannequin from before.
  300. 9:27But this didn't happen all at once
  301. 9:29either.
  302. 9:31Feet floated off the floor, arms
  303. 9:33disappeared into the chest when crossed
  304. 9:35, the phone hid behind the head, and it
  305. 9:37froze like ice upon arrival.
  306. 9:40I fixed each and every one of these
  307. 9:42issues and put all those methods into
  308. 9:44the kit.
  309. 9:46You just need to specify it as a scene
  310. 9:48of sitting and answering the phone.
  311. 9:52And I’d like to brag a little here. The
  312. 9:54Samaneun case I’m working on now looks
  313. 9:56better than the original Korean channel
  314. 9:58or the original foreign channel Punn. It
  315. 10:01turned out visually excellent.
  316. 10:03Since Samaneun is a case with many real
  317. 10:06victims, I created the scenes using only
  318. 10:08sets and lights. There are five secrets
  319. 10:10to it. First, I built the entire case as
  320. 10:13a village model on a black pedestal.
  321. 10:16When the pin light clicks on, the
  322. 10:17nameplate is revealed, and the
  323. 10:19criminal's path is shown only by a
  324. 10:21single red thread.
  325. 10:22Instead of people, the thread and window
  326. 10:25lights tell the story.
  327. 10:27Second, I built a separate, complete
  328. 10:29stage for every location. From the
  329. 10:31police sub-station and post office
  330. 10:32exchange to the investigation board and
  331. 10:34the bridge.
  332. 10:35I even set different lighting and colors
  333. 10:37for each location.
  334. 10:38I also created each prop one by one.
  335. 10:41The switchboard, the magnet-style phone
  336. 10:43, and even the police jeep.
  337. 10:45The shapes were generated with a free AI
  338. 10:47that turns a single image into 3D, and
  339. 10:48Claude refined them by applying
  340. 10:50materials in Blender.
  341. 10:52Even to my eyes, the prop detail is
  342. 10:54definitely better. Third, it's the
  343. 10:56realistic human movement you saw earlier
  344. 10:57.
  345. 10:58By adding a police hat and a
  346. 11:00mustard-colored uniform, it's
  347. 11:01immediately clear who the officer is.
  348. 11:04Fourth, there are no cuts; it moves
  349. 11:06continuously. The camera keeps rotating
  350. 11:08.
  351. 11:09The lights dim according to the story,
  352. 11:11and a clock ticks away at the bottom of
  353. 11:12the screen.
  354. 11:14If you measure the on-screen movement
  355. 11:16against that original Korean channel's
  356. 11:18first video, it moves about 1.5 times
  357. 11:19more.
  358. 11:20Fifth, I paired actual footage with 3D
  359. 11:23elements. I show a brief clip of the
  360. 11:25news from that time and transition
  361. 11:27directly into a model with the same
  362. 11:28composition.
  363. 11:30The screen you just saw was created
  364. 11:32without using a single paid AI video
  365. 11:34tool. I will update these methods into
  366. 11:36the kit as soon as they are organized.
  367. 11:38Once the video is up, please compare the
  368. 11:40original Korean channel with mine.
  369. 11:43So, what do you need to install on your
  370. 11:45computer to make videos like this?
  371. 11:47You only need three things.
  372. 11:50The first is Blender. It’s a 3D program
  373. 11:52used even by film studios, and it is
  374. 11:54completely free.
  375. 11:56Just go to the official Blender website
  376. 11:58, click the download button, run the
  377. 11:59file, and keep hitting next.
  378. 12:02I used version 4.2. Next, you install an
  379. 12:05add-on that creates human figures.
  380. 12:08It’s called MPFB; you just type the name
  381. 12:10into the extension menu inside Blender
  382. 12:12and click install.
  383. 12:14Using this to build a human figure and
  384. 12:16applying a clay-like texture gives you
  385. 12:17the faceless doll you saw earlier.
  386. 12:20But you don't need to install these one
  387. 12:23by one yourself.
  388. 12:26If you put the kit I mentioned into
  389. 12:27Claude Code or CodeX and type "install,
  390. 12:29" it checks if Blender is there and
  391. 12:31installs what is needed automatically.
  392. 12:34When you run it for the first time, it
  393. 12:36asks a few questions. What your channel
  394. 12:38name will be, which PC audio voice to
  395. 12:40use, and what the screen colors should
  396. 12:41be.
  397. 12:42Once you answer these, your videos will
  398. 12:44be generated with your channel's
  399. 12:46settings. I've set the default colors to
  400. 12:49the teal and amber I picked earlier. I
  401. 12:51recommend changing them to your own
  402. 12:53channel's unique colors. After all,
  403. 12:54color is what makes people recognize
  404. 12:56your channel just by looking at a
  405. 12:58thumbnail.
  406. 12:59It also checks your graphics card at
  407. 13:01this time, and if you don't have Nvidia
  408. 13:03, it guides you to skip the next steps.
  409. 13:06For the second part, Motion Bricks, you
  410. 13:08definitely need an Nvidia graphics card
  411. 13:10.
  412. 13:10It runs on Nvidia-exclusive acceleration
  413. 13:12technology.
  414. 13:14Nvidia stated it was tested on an RTX
  415. 13:164090, but my computer, which is a
  416. 13:19lower-spec RTX 4070 12GB model, ran it
  417. 13:21just fine.
  418. 13:23Those without that graphics card don't
  419. 13:24need to be disappointed.
  420. 13:26The waiting and looking-around motions
  421. 13:27we saw earlier were made using only the
  422. 13:29motion data I'll show you next, without
  423. 13:31Motion Bricks.
  424. 13:32Those users can simply move directly to
  425. 13:34step three.
  426. 13:37So, what’s the difference between using
  427. 13:38just the motion data and using it with
  428. 13:40Motion Bricks? Let’s compare the same
  429. 13:42scene.
  430. 13:44This is a scene where a detective enters
  431. 13:45through the door and sits in a desk
  432. 13:46chair.
  433. 13:48The motion data is a recording of an
  434. 13:50actual person acting it out.
  435. 13:52So, the number of steps and the stopping
  436. 13:54point are exactly as recorded.
  437. 13:56It doesn't know where the chair is
  438. 13:58located on our stage.
  439. 14:00If you just stitch the walking motion
  440. 14:01and the sitting motion together, the
  441. 14:03character will suddenly pop to the new
  442. 14:04position.
  443. 14:06So, when using only motion data, you
  444. 14:08have to cut from the walking scene to
  445. 14:10the sitting scene.
  446. 14:13If you make a cut like this, it still
  447. 14:15looks natural enough.
  448. 14:18Most stationary gestures can be done
  449. 14:20with just this.
  450. 14:21If you use Motion Bricks, you just
  451. 14:23provide the destination, like in front
  452. 14:25of that chair.
  453. 14:28Then, it automatically adjusts the steps
  454. 14:30to stop right in front of the chair and
  455. 14:31seamlessly continues into the sitting
  456. 14:33motion. The character moves within a
  457. 14:35single scene without needing a cut.
  458. 14:39To summarize, stationary gestures work
  459. 14:41with just motion data, but you need
  460. 14:43Motion Bricks to make them walk around
  461. 14:44the stage.
  462. 14:46You need two tools for installation. One
  463. 14:48is a tool for downloading large files,
  464. 14:50and the other is for setting up Python
  465. 14:51packages at once. The kit checks for
  466. 14:53these two, and installs them if they are
  467. 14:55missing.
  468. 14:56Then, when you download Motion Bricks
  469. 14:58and run it for the first time, it takes
  470. 14:59a few minutes to prepare for your
  471. 15:01graphics card.
  472. 15:02Do not turn it off thinking it has
  473. 15:04frozen. From the second time on, it will
  474. 15:06turn on quickly. Once it's running, the
  475. 15:07person will walk to the destination and
  476. 15:09stop like this.
  477. 15:10The kit also handles transferring this
  478. 15:11to the Blender character. There is one
  479. 15:13thing you must keep in mind.
  480. 15:15You can use the Motion Bricks code and
  481. 15:16AI model commercially, but the example
  482. 15:18motion files included inside are only
  483. 15:20allowed for testing.
  484. 15:22You should not use them as-is in your
  485. 15:23videos. So, for public videos, please
  486. 15:26use the motion data I will show you next
  487. 15:28; the third one is the motion data,
  488. 15:30Boneseed.
  489. 15:31It is available on an AI data site
  490. 15:33called Hugging Face. You cannot download
  491. 15:36it immediately; you must create an
  492. 15:37account and apply for access. When you
  493. 15:39apply, terms and conditions will appear
  494. 15:40, so make sure to read them yourself.
  495. 15:42If you earn revenue from YouTube, it is
  496. 15:44safer to treat yourself as a business
  497. 15:45and check the revenue criteria.
  498. 15:47Small businesses with annual revenue
  499. 15:49under $ 1 million can use it, but if you
  500. 15:51exceed that, you must pay a separate fee
  501. 15:53.
  502. 15:53And if you download this entire dataset
  503. 15:55, it is about 45GB.
  504. 15:57You don't need to download all of it. I
  505. 15:59have created and included a tool in the
  506. 16:01kit to download only the necessary
  507. 16:02motions.
  508. 16:03If you say something like "sitting and
  509. 16:04talking on the phone," it will find and
  510. 16:06download only that motion.
  511. 16:08Also, when you upload a video made with
  512. 16:10this data, you must write "Motion data
  513. 16:12by Bones Studio" in the description and
  514. 16:13include the link to the Bones Studio
  515. 16:15website.
  516. 16:17That is all for the installation and
  517. 16:19usage.
  518. 16:20Shall we see if a detective actually
  519. 16:22walks in if we give a one-line command?
  520. 16:25I typed just this one line into Claude
  521. 16:27Code.
  522. 16:29Make a scene where a detective walks
  523. 16:30into an office, sits in a chair, and
  524. 16:32answers the phone.
  525. 16:34Since it's my first time using it, it
  526. 16:35asks for my channel name first.
  527. 16:37Then, Claude builds the office set,
  528. 16:39places the detective figure at the door
  529. 16:41, and selects and stitches together the
  530. 16:44walking, sitting, and phone-answering
  531. 16:46motions.
  532. 16:47This is the result.
  533. 16:49He walks through the door, stops at the
  534. 16:51desk, sits in the chair, and answers the
  535. 16:53phone.
  536. 16:54Here, I'll add just one more instruction
  537. 16:56like a director.
  538. 16:57Have the camera slowly orbit the figure
  539. 16:59and keep the light focused only on the
  540. 17:01desk. But this time, it got too dark,
  541. 17:04and the detective wasn't very visible.
  542. 17:07Well, you can just fix that with words
  543. 17:09too. It's too dark. Add another light on
  544. 17:11the detective's face.
  545. 17:13Hold the phone with the left hand.
  546. 17:15If you say this, it just reshoots that
  547. 17:17scene.
  548. 17:18And just like that, the face is visible
  549. 17:20, and he answers the phone with his left
  550. 17:22hand.
  551. 17:23I didn't touch a single line of code
  552. 17:25myself.
  553. 17:28The reason this works is because Claude
  554. 17:30manages scenes in a table format.
  555. 17:32Each scene has fields for stage,
  556. 17:34character, motion, camera, and light, so
  557. 17:37when you edit with words, only those
  558. 17:39cells change while the rest remain the
  559. 17:41same.
  560. 17:44At first, my figure just stood there in
  561. 17:45a pose, but now he walks in, sits down,
  562. 17:47and even answers the phone.
  563. 17:50Because the character moves, it finally
  564. 17:52starts to look like a story is unfolding
  565. 17:53.
  566. 17:54There are drawbacks, too.
  567. 17:56First, Motion Bricks only runs on Nvidia
  568. 17:58graphics cards.
  569. 18:00If you don't have one, you have to stick
  570. 18:02to motion data, and walking sequences
  571. 18:03must be handled with cuts.
  572. 18:06Second, small movements like those of
  573. 18:07fingers or mouths are still awkward.
  574. 18:10I tried adding a smoking motion, but
  575. 18:12since the figure lacks fingers and a
  576. 18:13mouth, it turned out clunky like this.
  577. 18:16Third, 3D requires running the computer
  578. 18:18for a long time. You have to keep the
  579. 18:20computer on for quite a while to finish
  580. 18:22one episode.
  581. 18:23And this is more of a line to respect
  582. 18:25than a drawback. Since you're dealing
  583. 18:27with true stories, you shouldn't portray
  584. 18:29victims 'real names, faces, or death
  585. 18:31scenes sensationally.
  586. 18:33I don't make death scenes at all and I
  587. 18:35don't use real faces on screen.
  588. 18:37That way, viewers can watch comfortably
  589. 18:39, and the channel can last longer.
  590. 18:42There was one thing that held me back at
  591. 18:43first. It was the difficulty of making
  592. 18:44it.
  593. 18:45By leaving everything from installation
  594. 18:47to filming to Claude using the kit, that
  595. 18:48one issue was largely resolved.
  596. 18:51So, here is my answer.
  597. 18:53If you enjoy true crime stories, it's
  598. 18:55worth starting now.
  599. 18:56You can start even if you don't have an
  600. 18:58Nvidia graphics card. If you do have one
  601. 19:00, you can even make the figures walk
  602. 19:01around.
  603. 19:02Channels like this have just started to
  604. 19:04emerge in Korea.
  605. 19:07And for those curious about the endings
  606. 19:08of the two previews you saw today, check
  607. 19:10out the Cold Case Model Room via the
  608. 19:11link in the description.
  609. 19:13I've uploaded the versions where the
  610. 19:15figures come to life.
  611. 19:17If you just copy it as is, you'll only
  612. 19:19end up as a knock-off of someone else's
  613. 19:20channel. People subscribe not for the
  614. 19:22technique, but for that channel's unique
  615. 19:24fun and personality. Choose your stage
  616. 19:26colors, character designs, and the
  617. 19:28stories you want to cover.
  618. 19:29Try changing the way you develop those
  619. 19:31stories as well.
  620. 19:32The framework can be similar, but your
  621. 19:34channel needs its own unique identity.
  622. 19:36To be honest, you can get perfectly
  623. 19:39decent results using only free tools.
  624. 19:42However, some scenes that really need to
  625. 19:43grab the viewer's attention felt a bit
  626. 19:45lacking. So, I’m testing ways to elevate
  627. 19:48those specific scenes, and I'll show you
  628. 19:50in the next video.
  629. 19:52You know, in the past, while watching
  630. 19:54YouTube, you’d think, "Oh, this is fun.
  631. 19:55"
  632. 19:56Even if you wanted to try it yourself,
  633. 19:57you often couldn't because you lacked
  634. 19:59the technical skills.
  635. 20:01But now, if you do a little research
  636. 20:02with AI, we’ve entered an era where
  637. 20:04almost anything is possible.
  638. 20:07I research these things for you first to
  639. 20:09help you do it with minimal time and
  640. 20:11cost.
  641. 20:14You can get the kit I showed today and
  642. 20:16more information by joining my group
  643. 20:17chat.
  644. 20:19Please enter through the link in the
  645. 20:20pinned comment.
  646. 20:22Oh, and
  647. 20:24The video you are watching right now was
  648. 20:26also edited entirely with AI from start
  649. 20:28to finish.
  650. 20:29I didn't even open an editing program to
  651. 20:31create this video.
  652. 20:33If you want to learn how to use AI like
  653. 20:35this, please subscribe to the Brock
  654. 20:37Youth channel.

About this transcript

This page contains the full transcript of 9일 만에 210만 뷰, 3D 사건다큐 무료로 만드는 법 (클로드 코드 + 블렌더) by 불혹청년, generated from the public captions YouTube serves with the video. The transcript has 3,862 words across 654 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.