seedance-prompt-forge

Build clean, structured, consistent video prompts for Seedance and other text-to-video models (Kling, Runway, Veo) from simple inputs — instead of hand-formatting every prompt by hand.

forge turns a few simple flags into a properly ordered, model-friendly prompt. You stop re-typing the same phrasing, stop forgetting the camera/lighting/style structure, and stop getting inconsistent results because every prompt was written slightly differently.

forge --subject "lone samurai" --setting "rain-soaked neon alley" \
      --shot low-angle --movement slow-push --lighting neon \
      --mood tense --style dark-anime --pacing slow-motion --aspect 21:9

→

Lone samurai, in rain-soaked neon alley. Dark moody anime aesthetic, muted palette, dramatic shadows. Low angle shot looking up at the subject, heroic and imposing. Slow dolly push-in toward the subject. Saturated neon lighting, magenta and cyan glow. Tense, suspenseful atmosphere. Dramatic slow motion. 21:9 ultra-wide cinematic aspect ratio.

Why this exists

Text-to-video models reward prompts that are structured and ordered — subject first, then style, camera, lighting, mood, pacing, technical specs. Writing that by hand every time is tedious and error-prone, and small wording differences produce inconsistent results across a project.

forge encodes that structure once. You provide the intent (samurai, tense, low angle); it handles the phrasing and ordering that models respond to. Same inputs always give the same clean output — which matters when you are generating dozens of shots for one project and need them to feel consistent.

Install

npm install -g seedance-prompt-forge

Or use it as a library in your own project:

npm install seedance-prompt-forge

Requires Node.js 18+.

Usage

Command line

forge --subject "a wooden cabin" --setting "misty pine forest at dawn" \
      --shot extreme-wide --movement crane --lighting blue-hour \
      --mood serene --style cinematic --aspect 16:9

See every available preset key:

forge options

Full help:

forge --help

As a library

import { buildPrompt } from "seedance-prompt-forge";

const prompt = buildPrompt({
  subject: "a street vendor",
  setting: "a busy night market",
  shot: "medium",
  lighting: "neon",
  style: "documentary",
});

Flags

| Flag | What it controls | Example values | |------|------------------|----------------| | --subject (required) | Who / what is in the shot | "lone samurai" | | --action | What the subject is doing | slow-walk, turn, fight | | --setting | Location / environment | "neon alley" | | --shot | Camera shot / framing | wide, close-up, low-angle | | --movement | Camera movement | slow-push, orbit, tracking | | --lens | Lens character | anamorphic, macro, telephoto | | --lighting | Lighting setup | neon, golden-hour, low-key | | --mood | Emotional atmosphere | tense, epic, eerie | | --style | Visual style | cinematic, dark-anime, cyberpunk | | --pacing | Shot pacing | slow, fast, slow-motion | | --aspect | Aspect ratio | 16:9, 9:16, 21:9 | | --extra | Any extra free-text detail | "volumetric fog" |

Every flag accepts a preset key OR free text. Unknown values pass straight through, so you are never boxed in by the presets.

How it works

The prompt is assembled in the order text-to-video models weight most heavily:

1. Subject + action + setting (leads, because early tokens carry the most weight)
2. Visual style
3. Camera language — shot, movement, lens
4. Lighting + mood
5. Pacing
6. Extra detail
7. Aspect ratio (technical, goes last)

The vocabulary lives in small, readable modules under src/modules/ — camera, lighting, style, motion. Editing or extending the presets is just editing a plain object.

More examples

See examples/EXAMPLES.md.

Contributing

The preset vocabularies are intentionally simple to extend. Open a PR to add shot types, lighting setups, styles, or motion presets — each lives in its own module file. Issues and suggestions welcome.

License

MIT — see LICENSE.