YouTube2Text

힉스필드는 끝났다? 같은 장면 0원으로 다시 만드는 미친 로컬AI ㄷㄷ — Transcript

by 돈버는 온라인자판기 · 1,505 words · 258 segments · language en · Watch on YouTube

Full transcript

  1. 0:00I created the same scene using both the
  2. 0:03paid AI Kling 4.0 and the free local AI
  3. 0:06ComfyUI.
  4. 0:07One cost over 20,000 won, and the other
  5. 0:10cost zero won for video generation. This
  6. 0:13video is a senior-themed short-form clip
  7. 0:15titled "Mom's Seat." It is a story about
  8. 0:18the emotions a mother experiences at a
  9. 0:20wedding. First, please look at the
  10. 0:22character's expressions and gaze.
  11. 0:25>> It's my daughter's wedding.
  12. 0:27Now, let's look at the movement of the
  13. 0:29daughter leaving.
  14. 0:33Finally, let's compare the mother's
  15. 0:34expression.
  16. 0:36>> Can I sit there?
  17. 0:38>> Which one feels more natural and conveys
  18. 0:40the emotion better to you?
  19. 0:43The left side was made with the paid AI
  20. 0:46Kling 4.0, which used about 2,040
  21. 0:49credits for 14 clips.
  22. 0:51What does this look like if we convert
  23. 0:53it to Korean won?
  24. 0:54Based on the actual price we paid to
  25. 0:56recharge, 1 credit is about 10 to 15 won
  26. 0:59.
  27. 1:00So, this one short-form video cost about
  28. 1:0220,000 to 30,000 won to produce.
  29. 1:05Looking at the breakdown calculated by
  30. 1:07the creator, about 45%was spent on
  31. 1:09remaking or refining the clips.
  32. 1:12When you aren't satisfied with the
  33. 1:14result, the costs for re-generation add
  34. 1:16up. The right side was made in a local
  35. 1:19environment running AI directly without
  36. 1:21external services, so the extra cost for
  37. 1:24video generation was zero won.
  38. 1:26This local version was made with LTX 2.5
  39. 1:28, and in a local environment, you can
  40. 1:31also run video models like MiniMax H3
  41. 1:33directly on your own computer.
  42. 1:36This means you don't have to pay for
  43. 1:38external services to generate videos on
  44. 1:39a per-clip basis. I will also explain
  45. 1:41the comparison conditions. The local
  46. 1:43version used the same starting image as
  47. 1:45the paid version, and used the same
  48. 1:47voice and subtitles. Computer,
  49. 1:49electricity, and some voice-related
  50. 1:51costs are not included in this "zero won
  51. 1:53."
  52. 1:54I will show you why this difference
  53. 1:56matters and why it eventually leads to
  54. 1:59the story of AI companies.
  55. 2:01There are so many more people watching
  56. 2:03short dramas and video content these
  57. 2:05days. According to a survey, the number
  58. 2:08of users of major domestic short-form
  59. 2:10drama apps increased 3.3 times in one
  60. 2:13year, from about 900,000 to 2.98 million
  61. 2:16. The market outlook is also significant
  62. 2:18.
  63. 2:18The industry expects the global
  64. 2:20short-form drama market to grow from
  65. 2:22about 22 trillion won this year to about
  66. 2:2535 trillion won by 2030.
  67. 2:27The trend of creating stories into
  68. 2:29videos using AI is also accelerating.
  69. 2:32However, just because the market is
  70. 2:33growing doesn't mean everyone is making
  71. 2:35money.
  72. 2:36That's because, in addition to
  73. 2:38production costs, expenses for
  74. 2:39advertising to people are also involved
  75. 2:41.
  76. 2:42If you upload one video a day for 30
  77. 2:44days, it costs about 600,000 to 900,000
  78. 2:46won per month. If you add the monthly AI
  79. 2:49subscription fees you're already paying
  80. 2:51to that, the burden grows even larger.
  81. 2:55If production and marketing costs are
  82. 2:56higher than the revenue the video earns
  83. 2:58, losses accumulate the more you make.
  84. 3:00Ultimately, a business must move toward
  85. 3:02a structure that reduces costs and
  86. 3:03increases profits to survive for a long
  87. 3:05time. If you can make the same video for
  88. 3:08less, that difference becomes monthly
  89. 3:11savings.
  90. 3:13Yet, many videos recommend signing up
  91. 3:16for paid AI video services like Pika or
  92. 3:19sponsored AI tools.
  93. 3:22But signing up doesn’t automatically
  94. 3:24create videos, and the fees keep
  95. 3:26recurring every month.
  96. 3:28As you saw today, you don't necessarily
  97. 3:30have to do that.
  98. 3:32However, installing AI in a local
  99. 3:34environment and asking it to "make a
  100. 3:36video" doesn't immediately yield the
  101. 3:38desired result.
  102. 3:40To run these models in ComfyUI, you have
  103. 3:43to connect nodes one by one and build
  104. 3:45the workflow yourself.
  105. 3:47Since even a single scene is tangled
  106. 3:49with so many nodes, it is truly
  107. 3:50difficult for a person to adjust them by
  108. 3:52hand every time.
  109. 3:54That’s why methods are emerging where
  110. 3:56trending AIs like GPT-6 Astra operate
  111. 3:59ComfyUI on your behalf. But every time
  112. 4:02you fix a scene, it consumes a large
  113. 4:04amount of tokens, so now those costs
  114. 4:06pile up instead. Because you have to
  115. 4:08match the character's face, movement,
  116. 4:10the emotion of the scene, and the flow
  117. 4:12between them.
  118. 4:14The left is the first local result
  119. 4:16before editing, and the right is the
  120. 4:18result after I provided feedback six
  121. 4:20times.
  122. 4:21We used the same starting image and
  123. 4:23audio, but the character’s movements and
  124. 4:25expressions have changed.
  125. 4:30>> May I sit there? The company saved that
  126. 4:34feedback as a reference document to be
  127. 4:36used in future projects. So that we
  128. 4:38don't have to explain it from scratch
  129. 4:40every time.
  130. 4:41That is why we delegated this complex
  131. 4:44work to AI employees. Employees inside
  132. 4:46the game are sitting there. But these
  133. 4:48employees actually work. I assigned them
  134. 4:51the task of finding products to use for
  135. 4:53shopping shorts.
  136. 4:55The AI employee researches candidates
  137. 4:57and reports the results along with files
  138. 5:00.
  139. 5:01Even without a person holding onto the
  140. 5:03nodes, once you assign tasks via the
  141. 5:04game screen, the AI employees handle the
  142. 5:07rest.
  143. 5:08As an AI-native company, we connected
  144. 5:10the entire process from planning to
  145. 5:13scene breakdown, creation, review, and
  146. 5:15revision. A great result is not just the
  147. 5:18product of a single model, but the
  148. 5:20result of managing this process. A
  149. 5:22regular company would have to assign a
  150. 5:24person to every one of these node tasks
  151. 5:26.
  152. 5:26Our company runs itself like an
  153. 5:28autonomous vehicle.
  154. 5:31So, at our company, there is no need for
  155. 5:34people to learn Comfy or use expensive
  156. 5:37AIs like GPT-6 Astra.
  157. 5:40It’s not just automation; we tested and
  158. 5:43refined the same scenes multiple times
  159. 5:45to produce actual results. A short-form
  160. 5:47factory that produces quality while
  161. 5:50automating. That is what makes us
  162. 5:52different from others. This method is
  163. 5:55not used only for senior storytelling
  164. 5:57videos.
  165. 5:58The shopping shorts and architecture
  166. 6:00shorts we made were created by the same
  167. 6:02AI company.
  168. 6:04This shopping short explained the
  169. 6:06history of masks through images and
  170. 6:08narration.
  171. 6:12It is a video assembled into a single
  172. 6:14story by adding movement to the screen
  173. 6:15and layering subtitles.
  174. 6:17The additional cost to create one
  175. 6:19shopping Short in this style was about
  176. 6:2130 won.
  177. 6:23This architectural video shows the role
  178. 6:25of earth retaining walls in the process
  179. 6:28of building an underground parking lot.
  180. 6:32Both cases differ in their production
  181. 6:34methods from the senior-focused video
  182. 6:35you just saw.
  183. 6:37The shopping video assembled images and
  184. 6:39audio, while the architectural video
  185. 6:41utilized scenes made with external video
  186. 6:43generation tools.
  187. 6:46Even if the materials and tools differ,
  188. 6:48the process of planning, creating scenes
  189. 6:50, reviewing, and refining continues.
  190. 6:52If you try using an AI video tool once,
  191. 6:55you get results quickly.
  192. 6:57But if you don't like it, you have to
  193. 6:59give new instructions, recreate it, and
  194. 7:01move results between multiple windows.
  195. 7:04As the number of videos increases, that
  196. 7:06work increases as well. We divided this
  197. 7:09process as if it were a company.
  198. 7:11The writer creates the story, the
  199. 7:13producer creates the scenes, and the
  200. 7:15reviewer checks the results.
  201. 7:17We made it so that everyone's results
  202. 7:20lead to the next task.
  203. 7:22Humans determine the direction and
  204. 7:24results, and AI employees record and
  205. 7:27connect those decisions to tasks based
  206. 7:29on that judgment.
  207. 7:31It becomes a standard for work that
  208. 7:32accumulates in the company, not just a
  209. 7:34one-time disposable answer.
  210. 7:36I view the way of using AI in four
  211. 7:38stages. The first is the stage of asking
  212. 7:40one AI a question.
  213. 7:42The second is the stage where humans
  214. 7:44directly connect multiple AIs.
  215. 7:47The third is the stage of creating an AI
  216. 7:49company with divided roles, and the
  217. 7:51fourth is the stage where the company
  218. 7:53checks its status while working and
  219. 7:55organizes what it has learned.
  220. 7:58Our AI company checks its operational
  221. 8:00status like a weekly health checkup and
  222. 8:02organizes unnecessarily long memories
  223. 8:05like going on a diet. We leave important
  224. 8:08knowledge and feedback so they can be
  225. 8:09used for the next task.
  226. 8:11If you only look at one result, the
  227. 8:13difference might seem small. However,
  228. 8:16the method of explaining from scratch
  229. 8:18every time versus building up standards
  230. 8:20and experience can result in a growing
  231. 8:21gap over time. This is precisely the
  232. 8:24structure I am building. But this
  233. 8:27structure didn't just end with
  234. 8:29automatically creating videos. It
  235. 8:31actually achieved high view counts as
  236. 8:34well. These are the view counts for my
  237. 8:36channel over the last 28 days. Among
  238. 8:37them, the videos with high short-form
  239. 8:40view counts are like this. A single
  240. 8:42short-form video generated about 1.8
  241. 8:44million views and an estimated revenue
  242. 8:47of about 1.21 million won.
  243. 8:49As you saw earlier, there are parts you
  244. 8:51can make for zero won, but that requires
  245. 8:53a certain level of computer hardware. So
  246. 8:56, this lecture contains the know-how to
  247. 8:58create short-form content at a cost of
  248. 9:00around 30 won, regardless of computer
  249. 9:02specifications.
  250. 9:04And I will show you the exact process of
  251. 9:06how AI employees handle research,
  252. 9:07creation, and even uploading without you
  253. 9:09having to click a single thing.
  254. 9:12The price of this lecture increases as
  255. 9:14time goes by.
  256. 9:15Check out the course details in the
  257. 9:17pinned comment now before the price goes
  258. 9:20up.

About this transcript

This page contains the full transcript of 힉스필드는 끝났다? 같은 장면 0원으로 다시 만드는 미친 로컬AI ㄷㄷ by 돈버는 온라인자판기, generated from the public captions YouTube serves with the video. The transcript has 1,505 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.