studio-minimax-h3-world-video

Community@sunset1896/studio-minimax-h3-world-video
Community skill — not reviewed. It was shared by 落嵛 and has not been checked by Neta. Read SKILL.md below before letting an agent use it.
Use it in Cohub

Paste this into your agent's chat — it fetches the skill and follows it.

Use the skill: studio-minimax-h3-world-video. Run `npx -y @neta-art/neta-search-cli@latest skills use @sunset1896/studio-minimax-h3-world-video --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.
Install from a terminal
npx -y @neta-art/neta-search-cli@latest skills use @sunset1896/studio-minimax-h3-world-video
Summary

Turn your world's story into a continuous video, with your own characters and places kept true to how you made them.

  • Reads your world's characters, locations, lore and finished works first, then works out the story, who is involved and what happens in order.
  • You get a story summary, a shot-by-shot plan and a list showing which character or place picture goes into each shot.
  • You approve each stage before anything is made, so you can change the plan or stop and keep the documents.
  • The final result is a video saved in your world, together with the shot plan and prompt that produced it.
  • Works best when characters and locations already have clear looks or images, and the story has a defined sequence of events.
SKILL.md
# Studio World Story H3 Video Skill

This skill treats the current Studio world as the source of creative facts and existing materials and Works as reference anchors. After story understanding, asset binding and user confirmation, it uses MiniMax-H3 to generate a video Work.

## Scope

Use this skill when the user's request involves at least one of the following:

- Making a video from the current world's story
- Extracting characters, plot, locations or story environments from the world
- Rewriting an existing world story into video shots
- Generating a MiniMax-H3 video with the world's characters and scenes
- Needing the H3 Prompt, shot table and asset bindings together

Hand these requests to other skills:

- Writing a video prompt from a single standalone description only: use the general video prompt flow
- Making only images, character cards or scene images: use the matching image or character skill
- Making a visual novel, text adventure or playable story game: use the AVG or game skill
- Ongoing chat with a world character: use the roleplay skill

## Core constraints

1. Analyze the story only after reading the world; a one-line user request does not justify skipping the world read.
2. By default, read the current world's worldConfig, all character materials, all location materials, other relevant materials and existing Works.
3. When the user names characters, storylines or scenes, organize around the named scope first, but still check their necessary relationship and environment dependencies.
4. World settings take strict priority. Character identity, appearance, relationships, event causality, location anchors and confirmed visual references must not be rewritten by the generation model.
5. Anything not defined in the world or the user's request is marked "undefined" and must not be passed off as a world fact.
6. Existing Works are creative output and also citable visual or story material; only media references that actually exist and are usable may enter the video generation input.
7. Character and scene assets must be bound through real reference inputs, not merely named in the Prompt.
8. Switching the video model applies only to the current generation flow: call MiniMax-H3 with `NETA_DEFAULT_VIDEO_MODEL=minimax-h3`, without permanently changing the world's default model.
9. Before generating, read the current video model's parameter declaration, pass duration and aspect ratio explicitly, and use only declared parameters.
10. Video generation is asynchronous; do not call the video finished before the generation result is confirmed.

## Stage routing

Run in this order and never skip a stage that is unfinished or unconfirmed:

1. `READING_WORLD`: read the world index, materials and Works, and build a world snapshot.
2. `MODELING_STORY`: sort out the story timeline, causal chain, character relationships, conflicts, turns, ending and unknowns.
3. `BINDING_REFERENCES`: bind characters, plot, locations and environment changes to real materials, Works and media references.
4. `ALIGNING_CREATIVE_BRIEF`: from the user's request, settle the video goal, shot range, duration, aspect ratio, audio and story fidelity.
5. `CONFIRMING_SHOT_PLAN`: output and confirm the story summary, binding list and continuous shot table.
6. `CONFIRMING_H3_SETTINGS`: confirm MiniMax-H3, resolution, aspect ratio, duration, audio and reference inputs.
7. `GENERATING_H3_VIDEO`: switch to MiniMax-H3 in the current call context and generate the Studio video Work asynchronously.
8. `REVIEWING_OUTPUT`: check story correspondence, character identity, environment anchors, shot continuity, aspect ratio, duration and audio state.

The detailed inputs, outputs, error contracts and verification methods for each stage are defined by these protocols:

- `references/world-reading.md`: world snapshot and media state reading.
- `references/story-analysis.md`: fact layering, storyline identification, timeline and causality.
- `references/reference-binding.md`: real bindings for characters, plot, locations, props and media.
- `references/h3-production.md`: MiniMax-H3 modes, parameters, Prompt compilation and the Studio generation call.
- `references/verification.md`: stage quality gates, current-world regression and release packaging checks.

Write stage output into text or media Works that stay viewable in Studio; do not leave long production content only in the chat.

## Stage confirmation and progression

After each stage, first show that stage's output and verification status, then advance through a Studio confirmation card. The card offers at least: continue, revise the current stage, and stop and export. Enter H3 settings confirmation only after the story understanding, binding list and shot table are confirmed; submit video generation only after the H3 settings and reference inputs are confirmed.

## Standard output

When the flow reaches creative confirmation, prepare these four outputs:

- World story understanding: the story facts used by this video, chronological order, causal relationships and unknowns
- Character/plot/environment binding list: the material ID, Work ID, path and media state for every generation element
- Story and shot table: continuous timeline, shot actions, camera, spatial anchors, sound and shot handoffs
- H3 structured Prompt: visuals, ambient sound and non-diegetic music organized as H3 requires

After the video is generated, deliver the H3 video Work in Studio and keep the Prompt, shot table and binding information that correspond to this generation.

## General failure handling

- World index or materials unreadable: stop at `READING_WORLD` and report what is missing.
- Conflicts between story sources: stop at `MODELING_STORY`, list the conflicts and do not rule on them yourself.
- No usable reference for a character or scene: stop at `BINDING_REFERENCES` and mark the missing reference.
- H3 parameters cannot support the request: stop at `CONFIRMING_H3_SETTINGS` and propose feasible duration, aspect ratio or reference input changes.
- Video generation fails or identity drifts: fix only the failed shot, keep the confirmed story, bindings and Prompt, and never silently splice in bad footage.

## Operating rules

- Follow Studio's confirmation card mechanism; submit no expensive video generation before a confirmation card.
- Reply in the user's language; story documents and binding lists use the user's language by default.
- The generation fields and timing format of the H3 Prompt follow MiniMax-H3's local reference specification.
- The detailed flow must not copy any specific world's characters, plot or locations; always read them from the current world at run time.
Related skills
create-comicCommunityTurn any story or character into finished comic pages, with panels, speech bubbles and text, or a whole serialized comic.studio-ttsCommunityMake your characters and narrators speak aloud in a voice that fits them, or turn recordings into text.Windows VM in StudioCommunityGet a real, working Windows desktop living inside your world, operated right from the browser.portable-debian-openshot-setupCommunitySet up a portable Debian system and a working OpenShot video editor, even on a locked-down machine without admin rights.