NotebookLM: Build ONE Faceless Video Engine That Works for ANY Niche — Transcript

Full transcript

1. 0:00Most faceless video engines have one
2. 0:02problem. They work great until you
3. 0:04change the niche. Move from history to
4. 0:07finance or science and suddenly you're
5. 0:09rebuilding the entire workflow around
6. 0:11new sources, writing rules, and visuals.
7. 0:14[music] So today, I want to build the
8. 0:16opposite. One faceless video engine
9. 0:18where the niche is just an input. I'll
10. 0:21build it around ancient history, but at
11. 0:23every step I'll show you what changes
12. 0:25for another niche and what stays exactly
13. 0:28the same. For the demo, we'll create a
14. 0:31short documentary about Gobecée,
15. 0:33the 11,500year-old
16. 0:36site that challenged how we think about
17. 0:38early complex societies,
18. 0:40research, script, cinematic scenes,
19. 0:43diagrams, animation, and the final edit.
20. 0:47I'll show you snapshots from the result
21. 0:49as we go and the complete video at the
22. 0:51end. But the point isn't to make one
23. 0:54video in the history niche. It's to
24. 0:56build a reusable engine you can use to
25. 0:58make many more in any niche. Let's start
26. 1:01with topic and script. For this demo,
27. 1:04the niche is ancient history, but the
28. 1:06niche is just a variable inside the
29. 1:08engine. Open notebook LM and create a
30. 1:12new notebook. I'm adding three broad
31. 1:14sources. [music] Ancient history on
32. 1:16Wikipedia, Civilization from World
33. 1:19History Encyclopedia, and Cradle of
34. 1:21Civilization on Wikipedia.
35. 1:24If you haven't narrowed your niche
36. 1:25further yet, broad sources are better at
37. 1:27this stage because we want the topic
38. 1:29engine to see as many directions as
39. 1:31possible before we choose one. Open
40. 1:34settings, chat, custom, and paste in the
41. 1:37topic engine prompt. Set the niche to
42. 1:40ancient history. Then type, give me 10
43. 1:43video ideas for my faceless YouTube
44. 1:45channel in ancient history. I'm choosing
45. 1:48[music] the 11,500year-old
46. 1:51temple that broke history. Gobeclete
47. 1:54along with the angle notebook LM gives
48. 1:56us. Now create a second notebook for
49. 1:59focused research. Click add sources.
50. 2:02Choose web. Keep fast research selected
51. 2:05and search goe excavations.
52. 2:08Notebook LM finds 10 focused sources. I
53. 2:12select all 10 and import them.
54. 2:15Now paste in the script engine prompt
55. 2:17and ask for a 750word script using the
56. 2:20title and angle from the first notebook.
57. 2:22That gives us a source grounded
58. 2:24narration. And these citation numbers
59. 2:26let me open the exact source behind each
60. 2:29claim. So I can verify the script
61. 2:31instead of simply trusting the generated
62. 2:33answer. For another niche, the sources
63. 2:36and writing rules change. The workflow
64. 2:38stays the same. Let me show you where
65. 2:41this is going. What you're seeing now
66. 2:43are snapshots from the final Gobecée
67. 2:45video created with this engine. The
68. 2:48script we just built will be broken into
69. 2:49cinematic scenes, diagrams, and motion
70. 2:52instructions, then produced in batches.
71. 2:55Next, I'll show you the visual planning
72. 2:57system that makes those final shots
73. 2:59possible. Now, we turn the script into a
74. 3:01visual production plan. Open chat GPT in
75. 3:04a new conversation. Paste the visual
76. 3:07plan prompt and replace the niche
77. 3:09variable with ancient history. The
78. 3:11prompt creates one global visual style,
79. 3:14then breaks the narration into five
80. 3:16columns. Scene, voice over, scene type,
81. 3:20image prompt, and motion prompt. There
82. 3:23are only two scene types. Cinematic
83. 3:25scene is for environments, people,
84. 3:27architecture, atmosphere, and historical
85. 3:30reconstruction.
86. 3:32Text/diagram
87. 3:33scene is for information that is clearer
88. 3:35as a map, timeline, comparison,
89. 3:38structure, or explanatory graphic. For
90. 3:41this documentary, I'm targeting 80%
91. 3:43cinematic and 20% text or diagram, but
92. 3:47that ratio is another variable. Finance
93. 3:50might use more diagrams. A more
94. 3:52cinematic story might use fewer. Every
95. 3:55row also gets its own image prompt and
96. 3:57matching motion prompt. So before
97. 4:00production begins, every line already
98. 4:02has a visual purpose. Change the niche
99. 4:05and the visual rules change. The
100. 4:07planning system stays the same.
101. 4:10Here's why that matters. If I reuse this
102. 4:13planner for finance, I don't rebuild the
103. 4:15table. I change the visual rules, more
104. 4:18charts, data callouts, comparisons, and
105. 4:21explanatory diagrams with fewer
106. 4:23cinematic reconstructions.
107. 4:25For productivity or AI, the same rows
108. 4:28might become interface shots, workflow
109. 4:30diagrams, before and after screens, or
110. 4:33process animations.
111. 4:35The planner still asks the same question
112. 4:37for every line. What is the clearest
113. 4:40visual for this idea? So, the niche
114. 4:42changes the visual language and the
115. 4:44scene mix but not the planning system
116. 4:46itself. If you want the topic engine,
117. 4:49script engine, visual planner, and flow
118. 4:52batch prompts from this video, comment
119. 4:54prompts below and I'll share the
120. 4:56download link. Now, step three is
121. 4:58production. Open Google Flow, create a
122. 5:01new project and use agent mode. In agent
123. 5:05settings, I set confirmation before
124. 5:07generating to never and keep both image
125. 5:09and video outputs at 16 to9.
126. 5:12For this demonstration, I'm only
127. 5:14producing the first five scenes. That's
128. 5:17enough to show the batch production
129. 5:19technique because those five already
130. 5:21include both types from the visual plan,
131. 5:23cinematic scenes and text/diagram
132. 5:26scenes. To finish the full video, there
133. 5:30is no new workflow to learn. You simply
134. 5:32repeat the same batch process for the
135. 5:34remaining scenes. Go back to the visual
136. 5:37plan and copy the image prompts for
137. 5:39scenes 1 through 5. In flow agent, I
138. 5:42type create these scenes. Then I label
139. 5:45scene 1 through scene 5 and paste the
140. 5:47matching image prompt under each one.
141. 5:50Send that once and the agent generates
142. 5:52five separate 16-9 images in one batch.
143. 5:56And this batch already shows both visual
144. 5:58types. Scenes 1, 3, and five are
145. 6:02cinematic, while scenes 2 and four are
146. 6:04text or diagram scenes. Now we animate
147. 6:07them. Copy the motion prompts for the
148. 6:10same five scenes.
149. 6:12In Flow, add the five generated images
150. 6:14to the agent prompt. Type [music]
151. 6:17animate these scenes and paste the
152. 6:19matching motion prompt for each scene.
153. 6:22And this is where the engine starts to
154. 6:24scale. There's no new production
155. 6:26decision every five scenes because those
156. 6:28decisions were already made in the
157. 6:29visual plan. Scene type, image prompt,
158. 6:33motion prompt, and voice over are all
159. 6:35sitting in one structured table. That
160. 6:38means changing from ancient history to
161. 6:40finance or AI mostly changes what goes
162. 6:43into the batches, not how the batches
163. 6:45are produced. After flow creates the
164. 6:48five animated clips, download the
165. 6:50finished clips. Before we make the final
166. 6:53edits and see the finished video, I
167. 6:55wanted to share something quickly. My
168. 6:57Faceless YouTube Engine guide walks you
169. 6:59through the complete nine-stage
170. 7:01production pipeline built specifically
171. 7:03for demonetization safe faceless videos.
172. 7:06Every chapter has a checklist, homework,
173. 7:09and a real sample video built from
174. 7:11scratch so you can follow every step.
175. 7:14Already making videos? This will 10x
176. 7:17your speed and quality. Brand new?
177. 7:20You'll have your first video done by the
178. 7:22end. Link in the description. Now, step
179. 7:25four is voice over music and editing.
180. 7:28First, open Google AI Studio. I'm using
181. 7:32Sadal Tiger, the knowledgeable middle
182. 7:34pitch voice because it fits this
183. 7:36documentary style. Copy the voice over
184. 7:39for the first five scenes from the
185. 7:41visual plan. Paste it into AI Studio.
186. 7:44Generate it. Check the delivery and
187. 7:46download the audio. For music, open
188. 7:49YouTube Studio and go to the audio
189. 7:52library. I filter genre to cinematic and
190. 7:55mood to dramatic. After previewing a few
191. 7:58options, I download How to Train Your
192. 8:00Dragonet by Ezra Lip. Now, open Cap Cut
193. 8:04and import the voice over, the five
194. 8:06scene clips, and the [music] music.
195. 8:08Then, place scene one through scene 5
196. 8:11over their matching narration using the
197. 8:13visual plan. If a clip runs longer than
198. 8:16its voiceover section, trim it so the
199. 8:18visual ends with the idea being spoken.
200. 8:21Add the music underneath and lower the
201. 8:23volume so it supports the narration.
202. 8:26Fade it where needed. I'm keeping the
203. 8:28edit clean. We do not need heavy
204. 8:30transitions between every shot. Clean
205. 8:33cuts are enough when the visuals and
206. 8:34story are already doing the work. Do one
207. 8:37final check of timing, audio levels, and
208. 8:40scene order, then export. Let's see the
209. 8:43result.
210. 8:44For generations, historians believed
211. 8:47civilization followed an unyielding
212. 8:49recipe.
213. 8:52First came agriculture, then permanent
214. 8:54cities, and only after those foundations
215. 8:56were laid did humans construct grand
216. 8:59monuments. [music]
217. 9:01But hidden beneath a lonely mountain
218. 9:03ridge in southeastern Turkey, a
219. 9:0511,500year-old
220. 9:07sanctuary shattered that historical
221. 9:09timeline completely.
222. 9:12Long before the invention of the pottery
223. 9:13wheel, written language, or metallurgy,
224. 9:15[music]
225. 9:19ancient hunter gatherers erected massive
226. 9:22stone circles that defied everything
227. 9:24scholars thought they knew about early
228. 9:25human society.
229. 9:28And that result is the proof of the
230. 9:30system. To make a video in another
231. 9:32niche, you change the sources, writing
232. 9:35rules, and visual mix, but you never
233. 9:37rebuild the engine itself. The same
234. 9:40structure carries each video from topic
235. 9:42selection all the way to the finished
236. 9:44result. And remember, I only built the
237. 9:47first five scenes to demonstrate the
238. 9:49system. For the full video, just repeat
239. 9:52the same batch process for the remaining
240. 9:54scenes. That gives you one reusable
241. 9:56faceless video engine for any niche. And
242. 10:00if you want to see another complete AI
243. 10:02workflow for creating faceless content
244. 10:04from start to finish, click the video on
245. 10:06your screen now. It breaks down another
246. 10:09full system you can use for your own
247. 10:10channel. I'll see you there.