time travel · публичный навык Open Skill

time travel

Give a name. You get a documentary-style explainer of that life: major deeds in order, voiceover, and one scene per deed. Runtime comes from the deed list, not a preset minute count. ## When to use - You want a full-life explainer from birth to the ending, told through the events that actually define this person - You have a name (or a short description) and want voiceover plus animation, not a slideshow of captions - You want to change the last film: rewrite narration, switch style, redo one scene, or remaster > Not the right tool when you need on-screen subtitles, a year-by-year chronicle, or a single unbroken long take across an entire life. ## How it works 1. Confirm the figure, then research and draft a major-deed list 2. You approve or edit the list (and pick language, likeness, and look) 3. Narration, locked character look, per-deed scenes, music, then a mixed film You will be asked: Chinese or English voiceover; use your reference portrait or generate a new look; cartoon/illustration or photoreal documentary. One style for the whole film. ## Example prompts **A well-known life** > Make a documentary about Leonardo da Vinci. Cover the major deeds of his life. **Chinese narration** > 用中文旁白讲解林则徐的生平，按主要事迹来，不要逐年流水账。 **Redo one part** > Keep the same look and deed list. Redo only the exile segment, then remaster. ## Invoke `/historical-figure-documentary`

@Dreamerss

ShortDrama

6 4 8

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3292461/2026-09-03/59vnpysb.jpg

# Historical Figure Life Documentary

## When to use
- The user wants a documentary-style life explainer: cover the life with **major deeds** (birth through death/ending), not a year-by-year chronicle.
- The user gives a historical figure's name (or a description) and wants voiceover plus animation to tell that life clearly.
- Also use when continuing or redoing part of the previous cut (rewrite narration, change style, redo one segment, remaster).

## Check for a redo first
If the user is editing an existing cut or work-in-progress: **reuse** the already confirmed language, likeness, visual style, and deed list. Do not ask again. Read `narration_script` back with `get_document`. Redo only the named step (narration → `update_document` + regenerate that segment or the full TTS; one shot → redo that keyframe and `clip_N`; then `mix_clips`). Re-ask an item only if the user explicitly wants to change language, likeness, style, or deeds.

## Core settings (ask on a new cut; reuse on a redo)
1. **Narration language**: Chinese or English. Do not assume a default.
2. **Protagonist likeness** (check whether preset asset `protagonist_ref` exists first):
   - Bound: options are "Use Image 1" / "Generate a new look".
   - Not bound: do not offer "Use Image 1"; offer "Generate a new look", or ask the user to upload a reference image.
   - If they choose "Use Image 1": run `recognize_image` first for appearance, age feel, and costume, and put those notes into later prompts.
3. **Visual style** (ask separately from likeness source): cartoon/illustration, or photoreal documentary. One style for the whole film. Do not mix.

## Hard output rules (non-negotiable)
1. **No subtitles on the finished film**: do not overlay captions, text, or burned-in type anywhere. Narration is voice only—no caption bars, no text in image/video prompts, and do not generate `.srt`/`.vtt` files.
2. **One deed, one scene; hold the current deed in each clip**: each clip shows only the deed being narrated. Use `i2v` with a slow move on that clip's keyframe. **Do not** use `fl2v` to morph into the next deed. Connect clips with the locked protagonist look, consistent grade, and continuous narration/BGM. See `references/flow-principles.md`; voice and mix details are in `references/voiceover-guide.md` and `references/mixing-guide.md`.
3. **Duration follows major deeds**: no fixed total runtime, no per-clip second cap. Extract **major deeds** in time order; **one major deed = one segment**. Weight: light = a brief bridge (time and place); medium = tell this one event clearly; heavy = cause, course, and impact. Video must match **that segment's actual voiceover duration**. Total runtime = sum of segments. Do not walk the calendar year by year, and do not drop a major deed to shorten the film.

## Steps
1. **Confirm the figure**: confirm the historical figure's name (if the user only gave a description, confirm the name first).
2. **Research the life and extract major deeds**: use `search_web` for biography (birth, upbringing, key events, achievements, ending/death). Build a **major-deed list** in time order; keep only events that define this person. For each row record: year/period, deed name, one-line point, weight (heavy / medium / light), age stage (child / youth / adult / elder). If search results are thin, say so and ask the user to fill gaps.
3. **Confirm the deed list**: show the table (including the estimated film shape) and use `set_guidance` with "Proceed with this list" / "I want to add or remove deeds". **Wait for confirmation** before writing or generating.
4. **Ask language, likeness, and style** (via `set_guidance`; skip on a redo when already known):
   - Narration language: Chinese / English.
   - Protagonist likeness: offer "Use Image 1" only if `protagonist_ref` exists; otherwise "Generate a new look".
   - Visual style: cartoon/illustration / photoreal documentary.
   Wait for clear answers before continuing.
5. **Write the narration script from the deeds*