Storyboard prompt sequencer for AI video
A film is many shots, not one long take — and AI video works the same way. The trick to a coherent multi-shot sequence is consistency tokens: an identical character and environment description repeated in every prompt. This sequencer manages those anchors for you and exports a clean, numbered prompt list.
The core problem: identity drift
Current AI video models generate each clip independently. If you write “a woman in a green parka” in shot 1 and “a woman in a green parka” in shot 5, the model may render two different women. The parka might change shade or style. The face will almost certainly differ. This identity drift is the most common failure mode in AI film projects, and it gets worse the more shots you have.
The practical fix is to make the character description longer, more specific, and byte-for-byte identical across every prompt. Even small rewordings — “green jacket” instead of “green parka”, “woman in her 30s” instead of “woman, 30s” — give the model enough latitude to drift. The sequencer solves this by storing the character anchor once and pasting it unchanged into every generated prompt.
How it works
You define two anchors once — a character description and an environment — then write a one-line action for each shot. The sequencer builds each prompt as:
[character anchor] + [environment anchor] + shot action + cameraBecause the anchors are byte-for-byte identical across shots, the model is far more likely to render the same person and place in each clip. You generate the shots one at a time and assemble them in any editor.
Structuring a shot list
Think of each entry as a single camera setup: what the subject does, where the camera is, and whether it moves. For example:
| Shot | Action description | Camera |
|---|---|---|
| 1 | Subject walks through door, looks around | Wide static |
| 2 | Subject’s face as she recognizes someone | Close-up push-in |
| 3 | Subject crosses the room toward the camera | Medium tracking |
| 4 | Subject reaches out her hand to greet | Over-shoulder, tight |
Each line stays narrow — one action, one camera move. Adding too much to a single shot produces motion confusion; leave the complex action to the assembly cut.
Tips for consistent sequences
- Be specific in anchors. “A woman, 30s, short red hair, green parka, pale complexion, small silver earrings” holds much better than “a woman”. Vague anchors drift.
- Vary action, not identity. Change what happens and the camera per shot; never reword the character description between shots.
- Keep shots short. 3–5 seconds each — short clips preserve identity and motion, and they cut together more cleanly than long ones.
- Match lighting and time of day. Add the same lighting phrase to the environment anchor so shots don’t jump between noon and dusk.
- Generate in order. Some models (particularly image-seeded generators) can use the last frame of shot N as the first frame of shot N+1. Even if yours does not, generating in order keeps your mental model clear.