5-Step Workflow: How to Make Viral AI Short Dramas (60M Views) — Transcript

Full transcript

1. 0:00This AI short drama hit 20 million views in just 24 hours.
2. 0:03And people recreating it get millions of views.
3. 0:06This type of content brings in a ton of collaborations,
4. 0:08followers, and clients who pay for AI videos.
5. 0:11So let me show you how to make a short drama like this,
6. 0:14from the very first idea to the finished edit,
7. 0:16all inside ChatGPT.
8. 0:18But first, let's watch the full episode.
9. 0:22It's all right. Take your time.
10. 0:24Next.
11. 0:26Thank you. Next.
12. 0:29Next.
13. 0:30Wait.
14. 0:31Haven't you already been in?
15. 0:33No, it's my first time.
16. 0:34Same face. All of them.
17. 0:37Next.
18. 0:39Next.
19. 0:40We still have 14.
20. 0:44Bring me a magazine, the latest issue. I need to look at something else.
21. 0:46Sorry. Accounting sent me up. I need a signature on…
22. 0:49Okay. Come here.
23. 1:03What's her name?
24. 1:04Uh, she's, she's new.
25. 1:06Should I fire her?
26. 1:10I'm really sorry. Forgive me. I'll pick it all up.
27. 1:11What's your name, my dear?
28. 1:14Lucy.
29. 1:15Lucy, a beautiful name.
30. 1:20Have you ever been photographed?
31. 1:22For my employee ID?
32. 1:25I want you to model for Soul.
33. 1:27Me?
34. 1:28Yes, Lucy.
35. 1:30I want you to be the face of it.
36. 1:34But I'm an accountant.
37. 1:35I've already heard that, darling. Where have you been hiding?
38. 1:38The third floor.
39. 1:40Get Luca.
40. 1:48Luca? Raphael's asking for you.
41. 1:50Yes, handsome. Eight o'clock. I remember.
42. 2:01Luca is here.
43. 2:02Our new face, my muse. Find her something beautiful to wear,
44. 2:06and do something with her hair, yes?
45. 2:13Luca.
46. 2:13Lucy.
47. 2:17Well then, let's begin, muse.
48. 2:22The workflow for this short drama is stupid easy,
49. 2:24I promise. And the main reason it's now hit over 60 million views is that it looks
50. 2:29ultra realistic.
51. 2:29Like a clip from an actual movie,
52. 2:32which isn't typical for AI short dramas.
53. 2:34That's gonna be your advantage, because all of these videos on social media right
54. 2:38now are just slop.
55. 2:40So, once you've connected Higgsfield to ChatGPT,
56. 2:42start a conversation with your idea.
57. 2:44We'll keep working in this chat as we build the episode.
58. 2:47For a short drama, pick a situation people understand quickly.
59. 2:51Here are three starting points.
60. 2:53First, a Cinderella story.
61. 2:55Someone ordinary gets an unexpected opportunity. Two:
62. 2:59Love against the odds. Two people want to be together,
63. 3:02but something stands in their way. Three:
64. 3:05A rivalry. Two people want the same thing,
65. 3:07and only one can get it. I'm gonna go with the first one.
66. 3:10Our story is about Lucy, an accountant at a fashion house.
67. 3:14She interrupts a casting, trips,
68. 3:16and catches the director's attention. He thinks she could become the face of the
69. 3:20house. But picking a familiar storyline is only the beginning. Next,
70. 3:24decide how long you want your video to be.
71. 3:26For a short drama like this, aim for one to three minutes.
72. 3:30Then build your story around these four parts that every short drama is built on:
73. 3:35The hook, the context, the escalation,
74. 3:38and the cliffhanger. I'll break all of this down with examples in every scene as we
75. 3:43generate the videos.
76. 3:43But there's one more thing a successful short drama needs:
77. 3:46memorable characters and a setting. Let's create those first.
78. 3:52For this, I'll be using Soul 2, the best image model for ultra-realistic faces and
79. 3:57locations, so we can avoid that typical AI face.
80. 4:01Okay, main character first. We'll name her Lucy.
81. 4:03I tell ChatGPT what I want Lucy to look like and ask it to generate a few options
82. 4:07with Soul.
83. 4:08I'm also describing the early-2000s lighting,
84. 4:10colors, and film texture, so she already fits the world of our story.
85. 4:15So here are a few versions of Lucy.
86. 4:17Now, avoid the ones where her face looks washed out.
87. 4:20I think I really like this one.
88. 4:22I give each reference a clear name,
89. 4:24in this case LucyFace, and ask it to use that reference when we generate the next
90. 4:29image. Use the same names throughout the chat so it's clear
91. 4:32which references belong in each scene.
92. 4:34Once you have several characters and scenes,
93. 4:36you don't want to accidentally use an older version.
94. 4:38Now let's give her that office-siren outfit.
95. 4:40I'm going to describe the clothes in the chat and ask for an outfit reference,
96. 4:44which I'll call LucyOutfit. You can also upload a reference image and explain which
97. 4:48details you want to use. And once you have the face and the outfit,
98. 4:52you need to combine them into one final character sheet that we'll use for the
99. 4:55video.
100. 4:55But there are a few rules for getting it right.
101. 4:58First, use a plain grey background rather than white or black.
102. 5:02Second, ask for three views.
103. 5:04Full body front, back, and a close-up of the face.
104. 5:07And lastly, erase the face from the full-body front view.
105. 5:10Otherwise, the video model may use the face from the close-up in one shot and the
106. 5:14full-body front view in another.
107. 5:15The goal is to give the video model one clear face reference to follow.
108. 5:18We will need two versions of Lucy,
109. 5:20one with her hair in a bun and her glasses on,
110. 5:23and another for after the fall with her hair down and no glasses.
111. 5:27I'll create the first sheet for my references,
112. 5:30then edit that sheet to make the second version.
113. 5:32To combine Lucy's face and outfit,
114. 5:34I'm using GPT Image because it works really well with references and preserves the
115. 5:38details.
116. 5:39Now I'm asking it to combine LucyFace and LucyOutfit into the first character sheet,
117. 5:43following those three rules. Yo,
118. 5:46she looks fire in this outfit.
119. 5:47So this is how Lucy looks at the start of the story.
120. 5:50Now let's make her second version.
121. 5:52For the second version, I'll ask ChatGPT to edit this sheet with Seedream 5.0.
122. 5:57Let her hair down, remove the glasses,
123. 5:59and keep her face and outfit the exact same.
124. 6:02Okay, Lucy is ready. I used the same workflow for the rest of the cast:
125. 6:06Raphael, the director; Rebecca,
126. 6:08his assistant; and Luca, the stylist.
127. 6:10And there's one extra tip I want to show you.
128. 6:12For small details like jewelry,
129. 6:14generate a separate close-up with Soul and attach it along with the face and outfit
130. 6:18references when you make the character sheet.
131. 6:20That's what I did for Raphael to get his necklace right.
132. 6:23And remember those models from the video?
133. 6:25For the casting models, I asked ChatGPT to generate the whole group together with
134. 6:29Soul.
135. 6:30Different faces, different poses, everyone dressed in black.
136. 6:33We can use that group image as a reference for the entire casting scene.
137. 6:39Now we need the location. This is the casting room inside a fashion house.
138. 6:43So I want a table for Raphael, a clear path for the models,
139. 6:46and space for Lucy to walk in.
140. 6:49I'm describing the casting room in the same chat and asking Higgsfield to generate
141. 6:52it. I want a three-quarter view so we can see the table,
142. 6:56the runway, and the space where Lucy enters.
143. 6:59A straight-on shot might look great,
144. 7:01but when you change angles, the video model has to guess what the rest of the room
145. 7:05looks like.
146. 7:05Showing more of the same layout helps keep the location consistent from shot to
147. 7:10shot. Now that our assets are ready,
148. 7:12I'm giving ChatGPT our story outline and asking it to build a scene-by-scene script
149. 7:15around the cast and location we've selected.
150. 7:17I want the dialogue, the action,
151. 7:20and the generation prompts for each scene. For this,
152. 7:23I attach two skills, both linked in the description.
153. 7:26The first one writes the script.
154. 7:28It has a hundred story ideas built in and picks the strongest one.
155. 7:33The second one handles the prompts.
156. 7:36Camera, lighting, and all the cinematic work.
157. 7:39Built specifically for Seedance,
158. 7:41so your generation win rate goes way up.
159. 7:44And look, now I've got the full script and a ready-to-use prompt for each scene.
160. 7:48Now let's get to the fun part and start generating.
161. 7:52The first scene has to cover two parts of the short drama,
162. 7:55the hook and the context. The hook should start with a peek of the scene.
163. 7:59The viewer sees the loudest moment of the story without knowing how it got there,
164. 8:03so they keep watching to find out the reason.
165. 8:05But when you get to the context, keep the emotional swings coming every 20 to 30
166. 8:09seconds. A starting point, then hope,
167. 8:12then another hit, then a chance.
168. 8:14Once I'm happy with the scene, I'm asking ChatGPT to generate it with Higgsfield
169. 8:18using Seedance 2.5.
170. 8:23It's all right. Take your time. Next.
171. 8:27Thank you. Next.
172. 8:29Next.
173. 8:31Wait, haven't you already been in?
174. 8:33No, it's my first time.
175. 8:34I swear I've seen you.
176. 8:36Next.
177. 8:37Same face, all of them.
178. 8:38There are a few things I want to fix here.
179. 8:41Remember, I promised to show you how to build your own short drama.
180. 8:44This matters because it also affects which shots we choose.
181. 8:48Let's start with the hook. We need this model's fall as our opening moment,
182. 8:51but right now it looks unrealistic and a little silly,
183. 8:55and the empty runway doesn't feel like a busy casting.
184. 8:58So I'm asking ChatGPT to rework that shot and add some models waiting in the
185. 9:03background. And listen to Raphael here.
186. 9:06It's all right. Take your time. Next.
187. 9:07He sounds American even though I asked for a French accent.
188. 9:10And also somehow Rebecca's headshot ended up with the model photos.
189. 9:14She's the assistant, so that makes no sense.
190. 9:16But the acting and camera work are pretty dynamic,
191. 9:19so I want to keep that. I'll fix the fall and the headshot by telling ChatGPT
192. 9:23exactly how I want those moments to play out.
193. 9:26For Raphael's voice, I need to describe what the accent actually sounds like.
194. 9:30I'm asking ChatGPT for a soft uvular R,
195. 9:34which is that French R. I'm not going to even try pronouncing it,
196. 9:37made at the back of the throat. So to get the accent right,
197. 9:40describe how it sounds down to the little details.
198. 9:47Ew, what is this?
199. 9:51It's all right. Take your time.
200. 9:53Next.
201. 9:55Thank you. Next.
202. 9:58Next.
203. 9:59Wait, haven't you already been in?
204. 10:02No, it's my first time.
205. 10:04Non. I swear I've seen you.
206. 10:07Next.
207. 10:12Same face, all of them.
208. 10:14Yeah, this works so much better.
209. 10:16I mean, look at her face. The acting is kind of crazy.
210. 10:20I also give each character a voice description and ask ChatGPT to carry it into
211. 10:24every scene where they speak.
212. 10:26Then I listen to the takes together to check whether the voices actually match.
213. 10:30So the first scene is done. We've got the hook,
214. 10:33the context of what's going on.
215. 10:35So now let's move on to scene two where the twist and the climax happen.
216. 10:39Here Lucy walks in with her paperwork,
217. 10:41trips, and catches Raphael's attention.
218. 10:43I'm asking ChatGPT to generate scene 2 using the script and the references we
219. 10:47selected. Let's see how it turned out.
220. 11:09So how do you create an unexpected turn like this in your own short drama?
221. 11:13Start with a feeling your audience can recognize.
222. 11:16For Lucy, it's: I don't belong here,
223. 11:18I just embarrassed myself, and now I'm gonna get in trouble.
224. 11:21But instead, she gets an opportunity.
225. 11:24That's what makes the situation relatable.
226. 11:26You don't have to work in fashion to recognize that feeling.
227. 11:28And you can use this idea in a completely different story.
228. 11:31Someone who usually goes unnoticed suddenly gets a chance to be seen.
229. 11:35So scene 2 is done. Now let me show you the biggest secret behind viral short
230. 11:39dramas. In scene 3, Raphael wants to find out who Lucy is,
231. 11:43so he gets up and walks over to her.
232. 11:45This is our step 3, the escalation.
233. 11:47This is where you need an unexpected turn.
234. 11:49New characters show up, the conflict grows,
235. 11:52things happen that the viewer just doesn't see coming.
236. 11:55So I describe both performances:
237. 11:57Lucy apologizes nervously, Raphael's boredom turns into curiosity.
238. 12:01I'm also spelling it out: he gets up from the red chair,
239. 12:05leaves the table, and crosses the floor toward her.
240. 12:15What's her name?
241. 12:16Uh, she's, she's new.
242. 12:18Should I fire her?
243. 12:24I'm really sorry. Forgive me. I'll pick it all up.
244. 12:25What's your name, my dear?
245. 12:29Lucy.
246. 12:30Lucy, a beautiful name.
247. 12:36Have you ever modeled?
248. 12:37All right, now let's make the next scene.
249. 12:39Here, Raphael makes the offer. I'm also asking ChatGPT to generate a separate
250. 12:43close-up of Rebecca calling Luca.
251. 12:46Have you ever been photographed?
252. 12:47For my employee ID?
253. 12:51I want you to model for Soul.
254. 12:53Me?
255. 12:54Yes, Lucy.
256. 12:56I want you to be the face of it.
257. 13:00But I'm an accountant.
258. 13:01I've already heard that, darling. Where have you been hiding?
259. 13:04The third floor.
260. 13:06Get Luca.
261. 13:14Luca? Raphael's asking for you.
262. 13:16Now, Luca arrives and Raphael introduces Lucy as his new muse.
263. 13:20Let's use the scene 5 prompt from our script and see how it turned out.
264. 13:24Yes, handsome. Eight o'clock. I remember.
265. 13:36What happened?
266. 13:38Our new face, my muse.
267. 13:41Find her something beautiful to wear.
268. 13:43And do something with her hair, yes?
269. 13:54Okay, there are a couple of things I want to change here.
270. 13:56Luca's entrance keeps jumping between front and back,
271. 14:00and those long pauses slow everything down.
272. 14:03We also lose Lucy's reaction because her
273. 14:06face has turned away. Now, to give Luca a proper entrance,
274. 14:09I want to build anticipation before we see his face,
275. 14:12and I want the dialogue to start while they're still moving instead of that long
276. 14:16silent walk around Lucy.
277. 14:18Luca can lift a strand of her hair and introduce himself.
278. 14:21We'll move closer so we can actually see how he reacts.
279. 14:24Now I point out what needs to change in this take and ask ChatGPT to update the
280. 14:29generation
281. 14:29prompt. Then we generate another version and compare the two.
282. 14:33Yes, handsome. Eight o'clock. I remember.
283. 14:43Luca is here.
284. 14:45Our new face, my muse.
285. 14:47Find her something beautiful to wear. And do something with her hair, yes?
286. 14:55Luca.
287. 14:56Lucy.
288. 14:57Okay, now it feels like we're getting to a whole new part of Lucy's story.
289. 15:01We're getting close to the ending, so this is where we want to pick up the pace.
290. 15:05The changing angles, Luca's entrance,
291. 15:08and all those little interactions make it feel like something big is about to
292. 15:11happen. Now, I only have one shot in the dressing room and I don't want to bother
293. 15:15with an extra location image.
294. 15:17So here's what we'll do instead. I'm asking ChatGPT to use our initial room as a
295. 15:22style reference and describe the new space in the prompt.
296. 15:24This is a 10-second scene. Luca brings her to the mirror,
297. 15:27tells her they're ready to begin, and we cut to black.
298. 15:30And this is where the cliffhanger comes in.
299. 15:32It's the final hook in a short drama that makes you want to watch the next episode.
300. 15:36We end just as we're about to get the answer we've been waiting for.
301. 15:40Maybe someone is about to make a decision,
302. 15:42reveal who they really are, or walk out with a completely new look.
303. 15:46Let's see how scene 6 turned out. Perfect.
304. 15:54Well then, let's begin, muse.
305. 15:59With this last scene, we leave people wanting the next episode. Let's watch it.
306. 16:06It's all right. Take your time.
307. 16:08Next.
308. 16:10Thank you. Next.
309. 16:12Next.
310. 16:14Wait.
311. 16:15Haven't you already been in?
312. 16:16No, it's my first time.
313. 16:18Same face. All of them.
314. 16:21Next.
315. 16:23Next.
316. 16:24We still have 14.
317. 16:27Bring me a magazine, the latest issue. I need to look at something else.
318. 16:30Sorry. Accounting sent me up. I need a signature on…
319. 16:32Okay. Come here.
320. 16:46What's her name?
321. 16:47Uh, she's, she's new.
322. 16:50Should I fire her?
323. 16:53I'm really sorry. Forgive me. I'll pick it all up.
324. 16:55What's your name, my dear?
325. 16:57Lucy.
326. 16:58Lucy, a beautiful name.
327. 17:04Have you ever been photographed?
328. 17:05For my employee ID?
329. 17:09I want you to model for Soul.
330. 17:11Me?
331. 17:12Yes, Lucy.
332. 17:14I want you to be the face of it.
333. 17:17But I'm an accountant.
334. 17:19I've already heard that, darling. Where have you been hiding?
335. 17:22The third floor.
336. 17:24Get Luca.
337. 17:32Luca? Raphael's asking for you.
338. 17:34Yes, handsome. Eight o'clock. I remember.
339. 17:44Luca is here.
340. 17:46Our new face, my muse. Find her something beautiful to wear,
341. 17:49and do something with her hair, yes?
342. 17:56Luca.
343. 17:57Lucy.
344. 18:01Well then, let's begin, muse.
345. 18:05So, congrats. Now you know how to make super realistic AI short dramas.
346. 18:10Pretty simple, right? And making them look cinematic gives you a huge advantage in
347. 18:14this market over all those sloppy AI videos.
348. 18:16But even those are getting views and making money.
349. 18:19So I looked into how creators monetize their short dramas and found three ways you
350. 18:24could turn this into paid work.
351. 18:27First, make short dramas for clients. Second,
352. 18:31paid brand integrations. Brands like to collaborate with short drama creators
353. 18:35because people keep watching for the story while getting familiar with the product
354. 18:39along the way.
355. 18:40That builds brand awareness and can help drive sales.
356. 18:44Third, sell access to the next episode.
357. 18:46You might be thinking, who's going to buy if you've just started a new page?
358. 18:50That's what the free episodes are for. People watch them,
359. 18:53get curious, and want to know what happens next.
360. 18:55People are already paying for this kind of entertainment.
361. 18:58You can use a creator platform to offer a monthly subscription and test a price,
362. 19:02like $5 or $10.
363. 19:04500 subscribers paying $5 would bring in $2,500 before fees and costs.
364. 19:10The challenge is building a series people want to keep paying for.
365. 19:13You'll find the Higgsfield plugin link below,
366. 19:15along with the prompts and references from this tutorial,
367. 19:18so you can try the workflow in ChatGPT yourself.
368. 19:20Post it, tag the Higgsfield Creators account,
369. 19:22and share your results in the comments.
370. 19:24And if you want to put those skills to the test,
371. 19:26come join our Discord contest.
372. 19:28We just hit 200,000 members, so we're celebrating with a contest with a prize pool
373. 19:33of 200,000 credits. The invite link is in the description,
374. 19:36and you'll find all the contest details in the server.
375. 19:38And as always,
376. 19:39if you find this video helpful,
377. 19:42hit the like button, subscribe.
378. 19:45Oh, and I'll see you guys in the next one.