How to Prompt a 12-Second AI Movie Trailer (Six Shots, One Photo)
The AI movie trailer trend looks simple from the outside: you upload a selfie and get back twelve seconds of yourself walking away from an explosion. Getting a video model to produce something that actually feels like a
The AI movie trailer trend looks simple from the outside: you upload a selfie and get back twelve seconds of yourself walking away from an explosion. Getting a video model to produce something that actually feels like a trailer, and not one long, wobbly shot, took a fair amount of prompt work.
We built four trailer templates at TalkPix: action, rom-com, horror and thriller. Each turns one photo (two for the rom-com) into a 12-second widescreen trailer with sound. This post is the prompt structure behind them and the rules we ended up with.
The skeleton
Every trailer prompt has the same shape: one line for the look, one line for who is in it, six timed shots, and a line for the sound.
Widescreen [genre] movie trailer, 12 seconds, six shots with hard cuts,
[the look: colour grade, light, lens].
The lead is [the person in the photo], same face and hair in every shot.
0-2s: [an opening shot: the face, or the place]
2-4s: [the genre's signature set piece]
4-6s: [movement]
6-8s: [escalation]
8-10s: [the twist, or the drop to black]
10-12s: [a close-up and one spoken line: '...']
No title text. Sound: [the score], [two or three effects], the one line.
Six two-second shots fit a 12-second clip and still give each beat room to read. Here is what that looks like in the rom-com version:
Six rules we ended up with
1. Put a timestamp on every shot. "0-2s", "2-4s" and so on. A trailer is defined by its pacing, and naming the second each beat starts on keeps the cuts where a trailer would have them, instead of leaving the model to decide how long each idea deserves.
2. Say "hard cuts". It asks the model to cut between scenes rather than blend them, and the cutting is what makes it read as a trailer.
3. Repeat the identity line. "Same face and hair in every shot." Every cut is a new shot the face has to survive, so the prompt says it outright. With two leads we go further: "the same two people with the same faces, hair and clothes in all six shots, and nobody else takes their place in any shot."
4. Ask for no title text. Video models still spell badly, and a misspelt title card ruins the joke. Leave titles out of the prompt and add them in an editor if you want them.
5. Write the sound, not just the picture. The model generates audio too, so each genre gets its own sound line:
- Action: a deep trailer braam, rising strings, the explosion, tyres.
- Rom-com: a bouncy upbeat score, cafe chatter, rain, laughter.
- Horror: a low drone, creaking wood, a heartbeat, silence, then a whisper.
- Thriller: a ticking clock, a pulsing low synth, rain, a glass set down.
6. End on one short line. The last two seconds are a close-up and a single spoken line. It gives the trailer a punchline, and a short sentence is all that two seconds leaves room for.
For horror there is one more line in the prompt: "suspense only: no blood, no injury". It helps keep the render clear of content filters, and a door opening by itself is scarier anyway.
The four trailers
| Genre | The six beats | The last line | Photos |
|---|---|---|---|
| Action | the look up, the fireball, the rooftop leap, the car, the dust, the close-up | "Let's finish this." | 1 |
| Rom-com | the coffee bump, the rain, the kitchen disaster, the airport run, the rooftop dance, foreheads touching | "Took you long enough." | 2 |
| Horror | the old house, the hallway, the breath, the mirror, the flashlight dying, the whisper | "Don't turn around." | 1 |
| Thriller | the wall of photos, the late reflection, your own number calling, the other you outside, fast cuts, the close-up | "What if it was me?" | 1 |
Picking the photo
The photo matters as much as the prompt. What helps:
- One clear, well-lit face. No sunglasses and no group shots. The model has to carry that face through six different lighting setups.
- For the rom-com, two separate photos. You don't need a photo of the two of you together.
- Only photos of yourself, or of people who have agreed to it.
Making one without writing the prompt
If you just want the trailer, each of these is a ready template on TalkPix: AI action movie trailer, AI rom-com trailer, AI horror movie trailer and AI thriller trailer. You add the photo and it renders a 12-second 16:9 trailer with sound. Each costs from 72 credits at 480p (156 at 720p), credit packs start at $5 and no subscription is required. The rest of the templates are on the AI trend templates page.
The rom-com and horror examples above were made from photos of AI-generated people; the action and thriller examples use photos of a member of the TalkPix team, with permission.
Disclosure: I work on TalkPix.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.