Turn Claude into a Faceless Video Engine — Transcript

Full transcript

1. 0:00Most AI generated faceless videos have
2. 0:02the same problem. The character keeps
3. 0:04changing, the setting feels
4. 0:06inconsistent, and the visuals look like
5. 0:08random images placed over a voice over.
6. 0:11With the faceless video engine we're
7. 0:13going to build today, those problems are
8. 0:15much easier to avoid. Here is an example
9. 0:18of what the engine can create. As you
10. 0:21can see, the character setting, color
11. 0:23palette, and visual style stay
12. 0:25consistent. The key idea is simple.
13. 0:29Build the visual system first, then
14. 0:31create the scene prompts. Here is the
15. 0:33process. First, Claude develops the
16. 0:36story and creates a consistent visual
17. 0:38system. Then, it turns the script into a
18. 0:41production board with the image and
19. 0:43motion prompts for every scene. We use
20. 0:46that plan to generate the visuals,
21. 0:48animate the scenes that need movement,
22. 0:50and assemble everything with voice over
23. 0:52and music. At the end, I'll show you the
24. 0:55finished video created with this exact
25. 0:57workflow. If you are into practical AI
26. 1:00video workflows like this, subscribe for
27. 1:02more. I share the complete process and
28. 1:05every prompt I use. Let's start inside
29. 1:08Claude. First, I open Claude, create a
30. 1:11new project, and name it late night
31. 1:13scroll video engine. Inside project
32. 1:16instructions, I paste the instructions
33. 1:18from the prompts file. These stay active
34. 1:21throughout the project, so I do not need
35. 1:23to repeat the same rules in every chat.
36. 1:26They tell Claude to use reliable
37. 1:28sources, avoid unsupported claims,
38. 1:30separate evidence from interpretation,
39. 1:33and keep the narration conversational.
40. 1:36All the instructions and prompts I use
41. 1:37are available in the description. Next,
42. 1:40I start a new chat and paste the web
43. 1:42research angles and script prompt. At
44. 1:45the top is a placeholder. For this
45. 1:47demonstration, I replace it with late
46. 1:49night scrolling, but you can use any
47. 1:51subject. Claude searches the web,
48. 1:54summarizes the strongest findings,
49. 1:56explains important limitations, and
50. 1:58provides citations so I can check the
51. 2:01factual foundation before turning the
52. 2:03research into a story. Then Claude
53. 2:05suggests three story angles. The first
54. 2:08is your brain doesn't know the
55. 2:10difference between a slot machine and
56. 2:12your phone. The second is the real
57. 2:15reason you won't go to sleep, it's not
58. 2:17the phone. The third is shown on screen.
59. 2:21Each angle includes the central idea,
60. 2:23hook, main points, and takeaway. I
61. 2:27choose the first by typing angle one.
62. 2:30Claude creates a five-part outline, the
63. 2:32hook, the machine behind the scroll,
64. 2:35your brain at night, what the research
65. 2:38actually shows, and breaking the loop.
66. 2:41Then it writes the full narration using
67. 2:43the cited research. I review the
68. 2:45structure and claims. Type approved and
69. 2:48Claude saves the final narration as
70. 2:50final script.txt
71. 2:52inside the project. Now the story is
72. 2:55locked before we start designing
73. 2:57visuals. Next, I return to the Claude
74. 3:00project, start a new chat, and paste the
75. 3:02visual system and production board
76. 3:04prompt. Claude first defines the visual
77. 3:07system, then uses those rules to build
78. 3:09the production board. That order matters
79. 3:12because we decide what the video should
80. 3:14look like before asking for dozens of
81. 3:16individual images. When Claude finishes,
82. 3:20I open the generated HTML file in the
83. 3:23right panel. At the top is our title.
84. 3:26Your brain doesn't know the difference
85. 3:28between a slot machine and your phone.
86. 3:30Now I expand the visual system summary.
87. 3:33Our recurring character is Knox, a
88. 3:36genderneutral flat vector silhouette
89. 3:38with no face, fixed proportions, and the
90. 3:41same charcoal navy body in every scene.
91. 3:44The summary defines the style,
92. 3:46environment, pallet, camera language,
93. 3:48and scene mix. Setting the pallet first
94. 3:51helps avoid the random AI image look and
95. 3:54makes the video feel intentionally
96. 3:56designed. Claude plans roughly 35%
97. 4:00character scenes, 40% diagrams, 15%
98. 4:04metaphors, and 10% typography.
99. 4:07Because this story explains a mechanism,
100. 4:09diagrams get the largest share, while
101. 4:11the mix keeps the visuals varied. It
102. 4:14also gives us character and environment
103. 4:16reference prompts. Their essential
104. 4:18details are carried into the relevant
105. 4:20scene prompts, making consistency easier
106. 4:22later. Each scene card includes the
107. 4:25voice over, image prompt, motion prompt,
108. 4:28and a marked produced checkbox. Scene
109. 4:31one is a character scene, while for
110. 4:32example, scene 4 is a diagram scene. At
111. 4:36the top, filters let me isolate
112. 4:38character, metaphor, diagram, or
113. 4:41typography scenes. Before I open an
114. 4:44image generator, the visual rules,
115. 4:46prompts, scene categories, and
116. 4:48production tracking are already
117. 4:50connected. Now I open art to create the
118. 4:53visuals. One reason I use open art is
119. 4:56that the latest image and video
120. 4:57generators are available in one place.
121. 5:00Under image you can see options such as
122. 5:03chat GPT nano banana crad
123. 5:08and gro. For this video I'm using GPT
124. 5:11image 2. [music] I return to Claude and
125. 5:14copy Knox's character reference prompt
126. 5:16from the visual system summary. The
127. 5:18identity rules are simple. Genderneutral
128. 5:21flat vector silhouette, no face, fixed
129. 5:24proportions, and the same charcoal navy
130. 5:27body. I paste the prompt into open art
131. 5:30with four generations selected. Even
132. 5:32when the differences are subtle, several
133. 5:35options help me choose the best pose,
134. 5:37proportions, and framing for a reusable
135. 5:39reference. I generate four clean
136. 5:42options, compare them, and choose the
137. 5:44strongest Knox image to use as the
138. 5:46reference whenever the character
139. 5:47appears.
140. 5:49For this demonstration, I'm creating the
141. 5:51first six scenes. They already include
142. 5:54character and diagram visuals so we can
143. 5:56see how the same system handles
144. 5:58different scene types without producing
145. 6:00the entire video on screen. For scene
146. 6:03one, I click the plus icon in Open Art
147. 6:06and select the Knox image as the
148. 6:08character reference. [music] Then I copy
149. 6:10scene 1's exact image prompt from
150. 6:12Claude, paste it into Open Art and
151. 6:15generate. While scene one is running, I
152. 6:18start scene two using the same process.
153. 6:21When scene one finishes, I compare the
154. 6:23options and choose the third result
155. 6:25because the full body is clearly visible
156. 6:27and the composition works best. For
157. 6:30scene two, I also choose the third
158. 6:32option because the clock is easiest to
159. 6:34read. This is why multiple variations
160. 6:37help. I am choosing the result that
161. 6:39communicates the scene most clearly, not
162. 6:42simply the prettiest image. Then I
163. 6:44continue through the remaining scenes
164. 6:46with the same workflow. Copy the prompt
165. 6:48from Claude. Generate several options
166. 6:51and select the version that communicates
167. 6:53the shot best. Scene four is a diagram
168. 6:56scene. I copy its diagram prompt and
169. 6:59generate the options. You can see the
170. 7:02character bodies being used as figures
171. 7:03inside the diagram while the same pallet
172. 7:06and line style remain consistent. I
173. 7:09continue through scene six, checking
174. 7:11that the character, pallet, and overall
175. 7:13style still feel connected. These six
176. 7:16scenes demonstrate the full production
177. 7:18loop. The remaining cards follow the
178. 7:20same process. This is why I'm using Open
179. 7:23Art here. [music] I can move from image
180. 7:25generation into video generation in the
181. 7:27same workflow. My affiliate link is
182. 7:30below if Open Art fits your process.
183. 7:33Now, let's add motion. I open the
184. 7:36selected scene 1 image in Open Art and
185. 7:38choose video reference. Then, I return
186. 7:41to Claude, copy scene 1's motion prompt,
187. 7:44paste it into Open Art, and generate.
188. 7:47While that clip is processing, I repeat
189. 7:49the same workflow for the other scenes I
190. 7:51want to animate. The still image already
191. 7:54establishes the character, composition,
192. 7:56pallet, and illustration style. The
193. 7:59motion prompt only describes how that
194. 8:02existing scene should move, which helps
195. 8:04reduce unwanted changes.
196. 8:06After three animated scenes are ready, I
197. 8:08open the first result. The movement
198. 8:11makes the shot feel more cinematic
199. 8:13without losing the original
200. 8:14illustration. Then I compare the
201. 8:16animated scenes together. The bigger
202. 8:19advantage is consistency. Knox still
203. 8:22reads as the same character and the
204. 8:24pallet, line style, and overall visual
205. 8:26language stay connected from shot to
206. 8:28shot. That consistency matters for
207. 8:31retention because the viewer can focus
208. 8:33on the story instead of adjusting to a
209. 8:35completely different visual style every
210. 8:37few seconds. The six demonstration
211. 8:40scenes are ready. So now I put the
212. 8:42sample together in Cap Cut. You can use
213. 8:44any editor you prefer. First, I open
214. 8:48Google AI Studio. I already selected a
215. 8:51casual middle pitch voice because it
216. 8:53works well for this storytelling style.
217. 8:56I copy the narration for the first six
218. 8:58scenes. Generate the voice over and
219. 9:00download it. Next, I open YouTube
220. 9:03Studios audio library for background
221. 9:06music. I filter by cinematic for genre
222. 9:09and calm for mood. Listen to a few
223. 9:11tracks and download the one that best
224. 9:13matches the late night atmosphere.
225. 9:16Back in Cap Cut, the voice over goes on
226. 9:18the timeline first [music] because it
227. 9:20gives me the structure for the edit.
228. 9:23Then I import the six animated clips and
229. 9:25keep Claude's production board open
230. 9:27beside the editor as the assembly guide
231. 9:30so I can match each visual to the
232. 9:31correct narration. I place each scene
233. 9:34under the matching narration and adjust
234. 9:36its duration. If a clip is slightly
235. 9:39short, I slow it down. With these clean
236. 9:42line-based visuals, subtle, slower
237. 9:45motion still looks natural and gives the
238. 9:47viewer time to read the frame. I keep
239. 9:49the edit simple with clean hard cuts
240. 9:52instead of unnecessary transitions.
241. 9:55Then I add the music on a separate track
242. 9:57and lower the volume so it supports the
243. 9:59atmosphere without competing with the
244. 10:01narration. After a final audio and
245. 10:03visual check, I export the sample. Now,
246. 10:06let's watch the finished video. [music]
247. 10:09You told yourself five more minutes.
248. 10:12That was 40 minutes ago. You're not
249. 10:15stupid and you're not weak. [music]
250. 10:17You know you have to be up in 6 hours.
251. 10:20And yet here you are, thumb moving, eyes
252. 10:24locked on a screen that's somehow more
253. 10:26interesting than [music] sleep. If this
254. 10:28happens to you more nights than you'd
255. 10:30like to admit, you're in very [music]
256. 10:32crowded company. Surveys of American
257. 10:35adults have found that the vast majority
258. 10:37of people sleep [music] with their phone
259. 10:38in the bedroom. And a large share say
260. 10:41they're browsing the internet or
261. 10:42checking apps within 10 minutes of
262. 10:44actually falling asleep. [music]
263. 10:46So before we go further, let's drop the
264. 10:49idea that this is some [music] rare
265. 10:51personal failing. It's closer to a
266. 10:53universal experience. If you want me to
267. 10:56continue this series with a different
268. 10:58niche or format, drop the word continue
269. 11:01plus your niche in the comments. It
270. 11:03could be fitness, history, business,
271. 11:06relationships, anything you want to see
272. 11:08turned into a complete video engine,
273. 11:10drop it below. And one more thing, this
274. 11:14workflow isn't limited to this type of
275. 11:15faceless video. Click the video on your
276. 11:18screen now to see me apply the same
277. 11:20video engine idea in a completely
278. 11:22different way. People really like that
279. 11:25one. I'll see you there.