# Prompting Context Pack — MiniMax H3

The context pack for any agent that writes generation prompts for MiniMax H3
(video with native stereo audio). Give the prompting agent both files; they
are the model's own writing guides, straight from the official repo
(`MiniMaxAI/MiniMax-H3/docs/`).

## Files

- **`VIDEO_PROMPT_WRITING_GUIDE_base_en.md`** — the base guide. How to build
  prompts for all four modes: T2VA (text→video+audio), I2VA (image→video),
  FL2VA (first/last frame), L2VA (loop). Structure, camera language, motion,
  audio/sound direction, dialogue handling, timing. Use for every generation
  prompt.
- **`VIDEO_PROMPT_WRITING_GUIDE_ref_en.md`** — the full-reference-mode rewrite
  guide. How the six rewrite sections are organized and written when a
  reference video/character is given. Use only for full-reference jobs.

## Rules of use

1. Read both guides before writing any H3 prompt — they define the model's
   vocabulary; prompts written outside it produce weaker generations.
2. Follow the base guide's structure for standalone generations; layer the
   reference-mode rules on top when the job is a rewrite.
3. Keep the guides verbatim — they are reference material, not editable copy.
4. When in doubt about a prompt's shape, restate the relevant section from
   the guide before drafting.

## Source

- https://huggingface.co/MiniMaxAI/MiniMax-H3/tree/main/docs