---
name: animate-film-scene-refs
description: "Recreate a film scene (up to 15 s) with your AI character in one MiniMax H3 Max reference-to-video request — the character sheet plus up to 8 film frames with the actor replaced by GPT Image 2.5, and the film's own audio; JSON prompt with a references block, prompt enhance disabled. Use when the user wants their character in a film scene from frames and sound, reference-to-video, \"Image 2 is shot 1\". 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-film-scene-refs · a film scene from frames + its soundtrack (H3 Max reference-to-video)

> 🏆 **Source quality recommendation:** Seedance 2.5 reference-to-video (`bytedance/seedance-2.5/reference-to-video`, references as @Image1… / @Audio1…, 4–30 s, native sound) is the supplier's preferred quality route. That recommendation is supported by the product-commercial comparison, not a measurement of every character task. The H3 methods below retain their own measured inputs, settings and timings. Use supported external tools; this is not a FlashBoards integration claim.

{film} · {scene} up to 15 s → one generation with the person's {character} in every shot. Tested on a 14.8 s, 5-shot scene → one 15 s generation.
Longer than 15 s → `animate-scene-pairs`. Same scene using a storyboard → `animate-storyboard-turbo`.

## Choose

| Priority | Settings | Time | Result |
| --- | --- | --- | --- |
| 🏆 best | `minimax/h3-max/reference-to-video` · 15 s · **768P** · `aspect_ratio: "21:9"` (the film's) · sheet + 8 frames + audio | 24.1 s generation, 34.7 s request | 5 of 5 shots in film order, character in every shot, sheet never on screen |
| drafts | same at **480P** | not run for this scene | tested only on 5 s filter tests |
| ⚡ fast take | → `animate-storyboard-turbo` | 9.0 s generation | 5 of 5 shots, same early cuts |
| 🏆 more pixels | 1080P (a refinement of native 768P video) | not tested | — |

## Limits of the request

Up to **9 images, 3 audio, 3 video** (12 files together); audio and video clips 2–15 s each, 15 s together; `duration` 5–15 whole seconds; `resolution` 480P / 768P / 1080P; `aspect_ratio` adaptive · 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16. In the prompt they are Image 1…9, Audio 1…3, Video 1…3.

## Steps

1. **Cut the scene into shots** and spend the 9 image slots: Image 1 = the {character sheet}; 8 frames — more for long shots (3 for a long first shot: beginning / middle / end), 2 for a shot with a turn, 1 for the rest.
2. **Replace the actor in every frame that shows the actor** — GPT Image 2.5 Sunburst edit (`openai/gpt-image-2.5/sunburst/edit`), quality **low**, picture 1 = the person's {character sheet}, picture 2 = the film frame, size 1920×800 for 2.4:1; one run each, no retries needed in the test. A frame with nobody in it stays untouched. Prompt (blind — no clothes named; saying that nothing of the actor's costume stays is what keeps even a small figure seen from behind: 4 of 4 on *Mad Max*, 9.10.2026, the same as the worded version):
   ```
   Picture 2 is a frame from a film. Shoot it again exactly the same — the same camera, framing, set, light, pose, head turn, gaze and expression — but the person in it is {character} from picture 1: their face, hair and clothes exactly as in picture 1. Nothing of the clothes, headwear or gear of the person in picture 2 stays. Seen from behind, it is clearly {character}. Change nothing else.
   ```
3. **Cut the film's audio** for the same window → Audio 1.
4. **Write the JSON prompt**: `output` ({film} and year, "N shots with hard cuts at the given times, 15 seconds"), a **`references` block**, `rules`, `characters` (role of the sheet, wardrobe and face taken from Image 1 — no looks spelled out), one entry per shot (`images`, `camera`, `action` with "Exactly one person is in the frame in this shot: {character}" and BEGINNING / THEN / END, `cut_at`), `sound`. Never describe the original actor. The references block (structure as tested; renumber to your shots):
   ```json
   {"references": {
    "Image 1": "only the look of {character}: their face, hair and clothes. It is a reference, it never appears on screen as a shot.",
    "Image 2, Image 3, Image 4": "shot 1, in this order: its beginning, middle and end",
    "Image 5": "shot 2",
    "Image 6": "shot 3",
    "Image 7, Image 8": "shot 4, its beginning and end",
    "Image 9": "shot 5",
    "Audio 1": "the soundtrack of the whole video from the first frame to the last: {what is heard, e.g. the music {character} plays}. Every {on-screen sound action} lands on a beat of Audio 1."
   }}
   ```
5. **Request** through your own supported external video tool, with the prompt and body settings:
   ```json
   {"duration": 15, "resolution": "768P", "aspect_ratio": "21:9", "prompt_expansion_mode": "disabled",
    "reference_image_urls": ["<{character sheet}>", "<frame 1>", "…", "<frame 8>"], "reference_audio_urls": ["<film audio>"]}
   ```
6. **Check**: shots in order, the character in each, the sheet never on screen, cut times (a scene-change detector on neighbouring frames).

## What to expect

- **Cuts land early**: asked 5.29 / 7.38 / 8.63 / 11.30 s, got 3.21 / 6.13 / 8.17 / 9.46 s (0.5–2.1 s early); the last shot got the rest (5.6 s instead of 3.5 s). Order held, timing did not.
- **Audio 1 is played back**, not re-created: correlation 0.89 with the reference at zero offset.

## Never

- A film frame with the real actor as a reference — the actor's face, hair and shirt come back (filter test, 5 s 480P). H3 does not refuse, it copies.
- The sheet as Image 1 without "only the look … never appears on screen" — the video opened with the sheet for the first second (filter test).
- `prompt_expansion_mode` balanced / quality on a scripted scene — see `animate-prompt-enhance`.
- Replacing a frame with nobody in it — it invents the character where the film had no one.
- A demo character's sheet, replace prompt or scene prompt for a person's job.

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

The final drum solo of *Whiplash* (2014), 14.8 s, 5 shots, with the demo character Mara Vale in every shot; Audio 1 = the drum solo ("every stick hit and cymbal crash on screen lands on a beat of Audio 1"). 7 of the 8 frames were replaced; the cymbal close-up (nobody in it) stayed untouched. The historical replace prompt named Mara's clothes; the generalized blind prompt above instead assigns picture roles. No local sample files are included.