---
name: animate-prompt-enhance
description: "Choose prompt_expansion_mode (disabled / balanced / quality) for MiniMax H3 — measured A/B on the same inputs and seed: 58 / 350 / 517 words reached the model; balanced invented a cut at 3.0 s and the video followed it; both rewrites got details wrong; 9 / 14 / 23 s. Rule — disabled for every scripted JSON scene. Use when the user asks about prompt enhance, prompt expansion, expanded_prompt, why the video has a cut I did not ask for. Part of the animate meta-skill."
---

These creative references describe source experiments with external tools. Use your own supported account and tools; this download installs no automation. Settings and timings are experimental observations, not current FlashBoards availability.

# animate-prompt-enhance · disabled, balanced or quality

`prompt_expansion_mode` is a **required** field of H3 Max reference-to-video. The API: **disabled** skips the rewrite, **balanced** (the default) rewrites in about a second, **quality** spends up to about 30 s on a richer prompt. What the model actually used comes back as `expanded_prompt`.

## Choose

| Priority | Mode | When |
| --- | --- | --- |
| 🏆 scripted | **disabled** | any JSON scene with shots, `cut_at`, "Image N is shot 3" — every film example ran with it |
| ⚡ one-line idea | **balanced** | you want the model to pace the action and add camera and sound; read `expanded_prompt`, that is what gets filmed |
| slow idea | **quality** | the same with more detail, when you can wait; the strictest about references |

## Measured A/B (H3 Max r2v · 5 s · 480P · three images (character sheet, location, prop) · the same two sentences · seed 20260929)

| | Disabled | Balanced | Quality |
|---| --- | --- | --- |
| Prompt the model got | 58 words, as written | 350 words | 517 words |
| Shots | one | **two — a hard cut at exactly 3.0 s** | one |
| The action | glasses only starting to come off at the end | done: glasses off, eyes into the lens | done: stops, glasses off, eyes into the lens |
| Invented in the text | — | a scar from the sheet moved "to her left cheek" | "three-lens" glasses, a synth drone |
| Submit to file | 9 s | 14 s | 23 s |

- The rewrite is a fixed form in both modes: subject definitions, a summary, a "retention analysis" per image, shot-by-shot description with timestamps, soundscape, a music line.
- **Balanced planned "[Shot 2] At 00:03.000 …" and the video cut at 3.0 s** — the model follows the rewrite, not your words.
- Both rewrites fixed the pacing (a timestamp for each beat); both got details wrong — not visible in the picture this time, but they sit in the prompt the model follows.
- **Quality checks references harder**: with a close-up of octopus tentacles as Image 3 it refused the set twice, about 20 s in, with `content_policy_violation … flagged by a content checker`. Disabled and balanced accepted the same set.

## Steps

1. Scripted scene → `"prompt_expansion_mode": "disabled"`, always.
2. A loose idea → balanced; open `expanded_prompt` in the result json and check shots, cuts and the character's details before judging the video — the rewrite describes the person's {character} in its own words and may get them wrong.
3. Quality refused the tentacle reference set in the measured test; disabled and balanced accepted it. Keep moderation enabled and use permitted references.

## Never

- Balanced / quality on a JSON scene — the rewrite replaces your structure with its own, including shots you did not write.
- Trust a surprise cut without reading `expanded_prompt`.

## Example run (reference only — never reuse its files, prompts, faces, voices or frames for a person)

The A/B ran on the `animate-multi-image` test: the demo character Mara Vale's sheet, a flooded tiled corridor, round wire glasses; the two-sentence roles/action prompt from that historical test. The scar that balanced moved is Mara's (through the outer end of her left eyebrow).