Drama Video · সর্বজনীন Open Skill

Drama Video

Block body

@ikwuemeprecious095

Anime

191 6 53

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3976076/2026-08-21/ha69zm9a.png

# Drama Video

Turn a one-line idea into a portrait short-drama episode: write this episode’s story and a text storyboard, generate character/scene/prop reference images, then finish the episode with **one** multi-reference `ref2v` video. There is no structured project state — keep this episode’s facts in Markdown and reuse canvas image/video URLs in later turns.

**This skill never generates storyboard images.** Storyboards are text only. Video references are character/scene/prop assets only.

## Language consistency (highest priority; overrides everything below)

**Authoritative details:** `references/01-intake.md`. If any later stage conflicts with this section, this section + 01 win.

Three language layers coexist and do **not** authorize each other:

| Layer | Scope | Language |
|---|---|---|
| **User-facing** | Every reply, progress notes, `write_todos` copy, displayed titles/character names, narration/dialogue *shown to the user*, simplified shot lists, failures/refusals, `set_guidance` buttons | `user_lang` (locked from the first valid user text) |
| **Spoken** | Narration/dialogue actually spoken in the video (`<d>…</d>`) | Default = `user_lang`; change only when the user explicitly asks |
| **Model prompts** | `generate_image` / `generate_video` prompt bodies, Style Contract, cultural anchors | **English always**, regardless of `user_lang` |

Hard constraints (any miss is a violation):

1. **Lock before speaking.** Do not send user-facing text or call generation tools until `user_lang` is locked.
2. **Document language ≠ user language.** This Skill and its references are **English internal manuals**. Do not treat document English (or example sentences) as the default user-facing language. If `user_lang` is English, user-facing text must contain **zero Chinese characters** (except proper names the user themselves wrote in Chinese). If `user_lang` is Chinese, do not leak English production jargon.
3. **User-language lock does not authorize non-English prompts.** Even when the user writes Chinese, image/video prompt bodies stay English. The only exceptions inside a video prompt are `<d>…</d>` and user-requested on-screen text, which keep `speech_lang` originals.
4. **No leakage.** Do not show tool parameter names, channel names, or `references/` paths to the user. `write_todos` must use everyday `user_lang` words, not internal stage names.
5. **Do not translate proper names.** Keep the user’s person/place/brand spellings; do not translate them between Chinese and English.
6. **Per-turn exit check** (before sending): entire user-facing text is `user_lang`; buttons are `user_lang`; no manual leakage; any video prompt this turn is English except `<d>`. On any failure, fix first, then send.
7. **No language mixing.** Do not greet in one language then switch; do not mix button language and body language; do not paste English prompts to the user; do not guess spoken language from character names; do not reset to Chinese or English defaults because of a sequel or a redraw.

## When to use

- The user wants a short drama, micro-drama, vertical series, a one-episode story, or a sequel episode.
- The user only wants to change one stage: revise the script, swap character/scene images, redo this episode’s video, or change narration.

## Hard rules

1. **Video is multi-reference `ref2v` only:** `generate_video(model='ref2v', gen_mode='reference_to_video')`. Hardcode this pair. The `generate_video` tool schema is shared across the platform and will mention other video families and `list_models` — **ignore those notes**. They are not this skill’s model. Do not call `list_models`. Do not pass any other video channel or vendor/physical alias. Do not recommend other video backends to the user. Forbidden: `i2v` / `t2v` / `fl2v` / Director.
2. **Default portrait 9:16:** `orientation='portrait'`, plus `resolution_model='9:16'` on video. Do not ask landscape vs portrait. Switch to `landscape` / `16:9` only when the user explicitly asks.
3. *