Lie Into Coffin · সর্বজনীন Open Skill

Lie Into Coffin

Lie Into Coffin

@ExploreFly

Internal

0 1 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/3h2wqgap.jpg

# Lie Into Coffin (ref2v single-shot)

Lock the character’s look from the user’s **reference image**, then generate one **5-second** clip from a **fixed storyboard**. Beat timing, camera, and motion live in `references/`; on each run **only swap the person** via `input_urls`—do not change the timeline.

## When to use

- User provides (or already has on canvas) a **character image** and wants a “lie into coffin / casket” video.
- User asks to **regenerate with another face or character**: reuse the same 5s script; change only the reference in `input_urls`.

## Reference files (progressive disclosure)

| File | When to read |
|------|----------------|
| `references/fixed_5s_coffin_plot.md` | **Before every generation** (fixed plot, second-level beats, ref2v prompt template) |
| `references/generation_contract.md` | Before calling `generate_video` (channel, duration, orientation) |

Do not invent beats from memory. Call `read_skill_file(lie-into-coffin, references/fixed_5s_coffin_plot.md)`, then follow `generation_contract.md`.

## Steps

1. **Confirm inputs**
   - Require at least **one character reference image** (attachment, canvas asset, or asset id in chat). If missing, ask the user to upload; **do not** call `generate_video`.
   - If orientation is unspecified, default **`orientation='portrait'`** (9:16). Use `landscape` only when the user asks for horizontal video.
   - Optional: `recognize_image` to verify a single subject with a visible face. **Do not** paste the full recognition text into the video prompt.

2. **Load the fixed script**
   - `read_skill_file` → `references/fixed_5s_coffin_plot.md`.
   - Build **one English** prompt from § “Ref2V master prompt”. **Do not** reorder or paraphrase timed beats; only fill `{WARDROBE_HAIR}` to match the reference outfit and hair.

3. **Generate video (one final clip)**
   - `read_skill_file` → `references/generation_contract.md`.
   - Call **`generate_video`** exactly as specified:
     - `model='ref2v'`
     - `input_urls` = **[user character reference asset_id or URL]** (first image = character; never put audio URLs here)
     - `duration=5`
     - `prompt` = assembled master prompt from step 2
     - `name`: short English label, e.g. `Lie into coffin - 5s`
     - `entity_type='final_videos'`

4. **Deliver**
   - Reply briefly in the **user’s language**: generated with the fixed 5s script; character from their reference.
   - On failure, report the tool error honestly; **never** fabricate video URLs.
   - On success, call `set_guidance` with suggestions such as: try another reference image, or switch to landscape.

## Collaboration

- **One-shot**: if a reference image exists and the user did not ask to confirm each beat, read references then call `generate_video`—do not re-negotiate the storyboard.
- **Content boundary**: ritual / fantasy / period **atmosphere** only—**no** gore, decay, abuse, or real-world self-harm encouragement. Refuse extreme gore requests and do not call generation tools.
- Optional `write_todos`: `Confirm reference image` → `Generate ref2v clip` (at most two items).

## Notes

- Video generation does not support NSFW; refuse if reference or prompt requires it.
- **Video prompt**: English only (see reference file).
- Do not call `mix_clips` (single 5s shot, no concat).