Video Prompt Director

# Superprompt · Video Prompt Director (expanded G72)

Paste into any agent. Fill the brackets. The output is ready to paste into Seedance, Veo, Higgsfield, or similar video models.

---

Turn this idea into production-ready prompts for `[VIDEO MODEL]`:

`[SCENE IDEA — e.g. "a night market in the rain, a courier weaves through the crowd, ends on a rooftop reveal of the city"]`

The rules that make video prompts actually work:

1. **One shot = one camera instruction, 15 seconds max.** Video models fail on compound moves. "Dolly in while panning left as the camera cranes up" produces soup. Split it into separate shots.

2. **Characters are described identically in every shot, word for word.** Write each recurring character's descriptor once ("a wiry courier in a yellow rain jacket, black cap, mid-20s") and paste it verbatim into every shot they appear in. Any variation and the model recasts the role.

3. **Each shot prompt contains, in order:** subject + one action in present tense, camera (shot size plus one move), lighting and mood, then 2–3 concrete style references — never just "cinematic".

4. **Continuity is your job, not the model's.** End each shot's description with the frame it should land on; start the next shot with a frame that can cut from it.

Deliver:
- The character descriptor block(s)
- A numbered shot list, each shot's complete standalone prompt in its own copy-paste code block
- One line per shot: the thing this model most often gets wrong for this type of shot, so I know what to check
- A fallback variant for the riskiest shot — one shot always fails

Don't pad prompts with "masterpiece, 8k, trending" incantations. Concrete nouns and one clear action beat magic words every time.