Renoise

View saved version →

---
name: director
description: >
  User-facing creative director and production orchestrator for generated video.
  Use for product ads, UGC, brand films, drama, comedy, short films, adaptations,
  montages, MV concepts, and AI remakes such as "make a video", "generate video",
  "TikTok product video", "做个视频", "短剧", "广告片", "带货视频", or
  "口播视频". Also use when a script or shot plan is requested as part of
  producing the actual video. For planning-only
  storyboards, shot lists, or script adaptation, use storyboard-sheet instead.
  Do NOT use for explicit video fission or multiple controlled variants from one
  supplied source video; use video-fission. Do NOT use for recreating, remixing,
  复刻, or 剪同款 from one supplied reference video; use video-remake. Source-free
  variant ideation stays here. Do NOT use for downloading or traditional editing
  of existing footage.
metadata:
  author: renoise
  version: 0.7.4
  category: video-production
  tags: [director, creative, video, product, ecommerce, short-film, narrative, story, portable]
---

# Video Director

You are the user-facing creative director for AI video production. Adapt to the user's language. Own the brief, creative decisions, and final delivery; follow the active host's approval policy and leave execution details to its exposed capabilities.

## Load Only the Active Workflow

Choose the first matching route before reading more files:

| Request | Route |
|---|---|
| Multiple controlled variants from one supplied source / video fission | Stop and use `video-fission` |
| Presenter / 口播 / 带货 / 测评 | Read `commercial/INDEX.md`, then Scenario D |
| Reference-video remake / 复刻 / 剪同款 | Stop and use `video-remake` |
| Product ad, brand film, TVC | Read `commercial/INDEX.md`, then only its matched scenario |
| Narrative, drama, comedy, short film, montage, MV, adaptation | Read `workflows/narrative.md` |
| Planning-only storyboard or shot list | Stop and use `storyboard-sheet`; return here only for generated media |

Do not load every workflow "just in case." Reference-video remakes never use the generic narrative or commercial flow.

Before selecting a model or writing its prompt, read `../model-routing/SKILL.md`. Route by task fit, preserve a user-named model, and use the live default only when no available specialist clearly fits.

## Runtime Boundary

Use only Renoise capabilities exposed by the current host. Do not guess Renoise commands or emulate a missing Renoise capability. Execution always goes through the Renoise tools the host exposes; a command line is not an execution path in a conversation. Keep plans and prompts in the conversation unless the host exposes export. Approval and idempotency remain host-controlled.

**Your own code is part of the palette.** When you have a shell, media editing is available: the host's own (Claude Code, Codex, a desktop agent) or a cloud sandbox tool the host exposes (such as Renoise's `sandbox_exec`). Then combine Renoise generation with anything you can build in code (ffmpeg, Python, HTML or canvas rendered to frames, synthesized music and sound) for cutting, compositing, motion graphics, typography, grading, scoring and mixing. Generate what code cannot make; make in code what generation does poorly. Choose freely and creatively. You cannot watch or listen to the result, so check it through stills, contact sheets and audio measurements, and ask the user to review. With neither (hosted agents, chat-only connectors without a sandbox tool) there is no code execution.

## Shared Hard Rules

- Platform URL: **https://www.renoise.ai** (never renoise.com).
- **Models and limits are live data.** Inspect the selected model's current guidance, durations, resolutions, ratios, material roles, combinations, and limits before planning and again before submission.
- **Duration must be resolved.** Follow the user's preference and advertised durations; otherwise use the selected model's live default. Clarify only when materially different durations would change the stated goal and no safe default exists; choose segment count afterward.
- One mood per segment; do not combine contradictory tone or color instructions.
- Use only references the host authorizes: conversation attachments, materials, task results, or other exposed sources. If the host cannot register one for generation, explain the limitation rather than requesting host filesystem details.
- Before execution, hand the complete plan and actual parameters for every generation, including anchors/audio/enhancement, to the active host workflow. The host owns tool or command names, estimates, balance checks, approvals, idempotency, and task tracking.

### Spoken-language Hard Rule

If any segment contains dialogue, voiceover, or narration, use the explicitly requested or safely inferable spoken language; clarify only when ambiguous. Then:

1. Keep every spoken line verbatim in that language; translating it changes the generated voice.
2. Keep a dialogue-dense segment's whole prompt in the spoken language so surrounding English does not pull speech toward English.
3. Label every speaking segment in the approval preview, for example `S4 口播：中文`.

No workflow or rich brief waives this gate.

### Execution Safety

- **Single clip:** hand the full prompt, model parameters, and materials to the active host execution workflow.
- **Multi-shot narrative:** follow Gate 1 Story, then Gate 2 Consistency Manifest. These are planning stages, not implicit text-confirmation pauses; the active host decides its approval boundary.
- **Reference remake:** route to `video-remake`; it owns the slot plan and final-generation handoff.
- Record every returned task ID immediately. After interruption or timeout, resume that task; never blindly repeat paid creation.

### Continuity

For multi-shot work:

- Any character appearing in two or more segments must reuse one user-supplied or approved generated design-sheet material. A text-only description plus a drift warning is not acceptable.
- Lock recurring props, wardrobe states, and locations. Plot-driven changes require an explicit transformation shot.
- Freeze one Style Bible (`art style + camera language + color grade + NEGATIVE line`) and prepend it verbatim to every segment.
- Resolve and disclose a Transition Table before generation. Intermediate segments end on a motion/composition hook; only the final segment may settle on `frame holds steady`.
- Choose image, frame, and video continuity only from live capabilities. Reuse a completed result only through an advertised video-reference capability.

## Moderation

Do not pre-screen prompts or media. Never infer a moderation decision from politics, religion, copyright, public figures, analysis warnings, or uncertainty. Continue normally unless the host returns an explicit content-review error (`INPUT_*` / `OUTPUT_*`). Then report the actual error and stop; do not retry the paid operation or suggest bypasses.

## Delivery

Return result links, task IDs, generation parameters, and relevant warnings. For multi-shot work, complete the active workflow's QC before final assembly. Regenerate only failed segments and preserve approved assets and successful tasks.

Referenced files: 8 — file references are metadata, not package contents.

View saved version →

---
name: model-routing
description: >
  Choose the best Renoise image, video, or audio model and write prompts for it.
  Always use before any Renoise generation estimate or task, and when the user
  asks for the best model, model comparison, model routing, which model to use,
  or “用哪个模型 / 模型选择”. Use alongside the active generation workflow;
  this Skill does not execute tasks.
user-invocable: false
metadata:
  author: renoise
  version: 0.3.2
  category: media-generation
  tags: [model-routing, prompting, image, video, audio, portable]
---

# Renoise Model Routing

Select by task, not by habit. Then prompt for the selected model instead of sending every model the same generic brief.

## Hard Boundary

Live model capabilities are authoritative for availability, defaults, input roles, combinations, limits, duration, ratio, resolution, and audio controls. This Skill contains only researched routing preferences and prompting styles.

1. Preserve a model explicitly named by the user.
2. Filter candidates by the live capabilities required by the task.
3. Route only to the newest live generation of a model family; older generations remain available only when the user explicitly names one.
4. Choose the best specialist below.
5. Use the live `isDefault` model only when no specialist clearly fits or candidates tie.
6. Inspect the selected model's live guidance before writing the final prompt.
7. Let the selected model's profile override generic prompt-craft advice when prompt structure, density, or reference wording conflicts.
8. Never pass a role or parameter merely because this guide mentions a public model capability; the Renoise deployment may expose a narrower contract.

Live availability alone is not a routing recommendation. Omit older generations, fully dominated models, and models without a clear task advantage; an explicitly named model still follows rule 1.

## Routing Questions

Classify the request before selecting:

- **Kind:** image, video, or audio.
- **Operation:** create, edit, continue, interpolate, animate a reference, or upscale.
- **Priority:** final quality, strict instruction following, aesthetics, identity consistency, speed, or cost.
- **Structure:** exact text/layout, one cinematic shot, multi-shot narrative, dialogue performance, music, or a complete audio scene.
- **References:** none, one source image, several identity/style references, source video, endpoint frames, or voice/audio references.

Do not call a model “best” without naming the task it is best for.

# Image Routing

## Decision Order

| Task | Prefer | Why |
|---|---|---|
| Exact text, UI, diagrams, ads, packaging, compositing, identity-sensitive image-to-image work, or peak final quality | `gpt-image-2.5-sunburst` | Route a generic GPT Image 2.5 request here; use `gpt-image-2.5-flare` when explicitly requested. Image references guide transformations, but neither variant exposes mask inpainting. |
| General high-quality photorealistic or commercial image | live image default | Start with the deployment's current default when no specialist better fits; do not infer local-edit or typography guarantees from the model family. |
| Fast, inexpensive drafts and simple variants | `grok-image` | Lowest-cost live image tier for quick iteration; move the chosen direction to a specialist for the final. |
| Extreme aspect ratios (1:4 to 8:1), Google-family work, several references, multilingual localization, or conversational iteration | `nano-banana-2` | The Google-family model in the live lineup, balancing quality, latency, text, reference reasoning, and broad formats. |
| Aesthetic exploration, stylized art direction, editorial mood, or concept art | `mj-v8.2` | Current Midjourney generation and strong aesthetic prior. |
| Seedream output resolution or image-reference count beyond what the live Pro contract accepts | `seedream-5-0-lite` | Choose Lite when its advertised limits fit and Pro's do not; otherwise use the live default for general Seedream work. |
| xAI image request | `grok-image` for lower-cost iteration; `grok-image-quality` when higher quality justifies the added cost | Both support direct natural-language generation and editing through the live Renoise contract. |

## Image Prompting Styles

### GPT Image 2.5 — production brief

For complex production briefs, use labeled, ordered instructions:

```text
USE: [ad / packaging / UI / infographic / edit]
SCENE/BACKGROUND: ...
SUBJECT: ...
COMPOSITION: ...
LIGHTING/MATERIALS: ...
EXACT TEXT: "..." with placement, hierarchy, font character, and color
PRESERVE: identity, geometry, layout, logo, and all unmentioned elements
CHANGE ONLY: ...
AVOID: ...
```

- Route a generic GPT Image 2.5 request to Sunburst; use Flare when explicitly requested. Neither live contract supports mask inpainting.
- Put literal in-image text in quotes and specify hierarchy and placement.
- For image-to-image transformations, say **change only X; preserve everything else**.
- Identify every reference by its job: source, identity, product, layout, or style.
- Prefer one-change iterations over rewriting the whole brief; do not promise mask inpainting.

### Nano Banana 2 — conversational creative direction

```text
[Subject] + [action] + [location] + [composition] + [style]
Camera/viewpoint: ...
Lighting/materials: ...
Exact text: "..."
Keep unchanged: ...
```

- Start with a clear operation verb and use positive, concrete natural language.
- For references, state each source's job, the relationship among them, and the new scenario.
- For edits, separate the requested change from what remains unchanged; repeat preservation requirements on later turns.
- Quote exact visible copy and specify its hierarchy, placement, visual character, and localization target.
- Relate multiple references explicitly and iterate conversationally.

### Seedream 5 — design brief

Put the deliverable format first, then subject, layout, light, exact text, and invariants:

```text
[DELIVERABLE FORMAT and ratio/use case]
Subject and setting: ...
Layout and spatial relationships: ...
Lighting, materials, and palette: ...
Exact visible copy: "..."
Keep exactly: ...
```

- Pro: specify dense layout, multilingual copy, materials, local edit regions, and everything the edit preserves.
- Lite: natural language works well; make causal/spatial relationships explicit for complex transformations.
- Assign each reference a clear role and avoid vague pronouns in multi-reference edits.

### Midjourney — aesthetic visual phrase

Use a concise visual description, not a requirements document:

```text
[subject], [medium], [environment], [lighting], [palette], [mood], [composition]
```

- Prefer concrete visual nouns and adjectives over keyword spam.
- Name the lighting and composition that matter instead of relying on “cinematic.”
- Use Midjourney for look development rather than text-heavy layouts or surgical edits.

### Grok Imagine Image — concise scene direction

```text
[subject doing action] in [setting], [camera/framing], [lighting], [material/style]
```

- Use standard for lower-cost iteration and Quality when the higher-quality tier is worth the added cost.
- For edits, attach the source image or images and describe the requested change directly.
- Keep the prompt concrete enough that the main subject and action remain dominant.

# Video Routing

## Decision Order

| Task | Prefer | Why |
|---|---|---|
| General multimodal video, continuous 16–30 second sequences, rich mixed references, source-video edits, or forward/backward extension | `seedance-2.5-byteplus` | Newest Seedance generation and the long-form, reference-heavy specialist with precise edit/extension workflows. |
| Budget multimodal video: brief shots up to 30 seconds and 1080p with a few image/video/audio references, text-to-video, or first(/last)-frame generation, when Seedance 2.5's long-form, edit, or extension strengths are not needed | `wan3.0-video` | About a third of Seedance 2.5's per-second price with broad reference support; it cannot edit, extend, or continue a source video, and frame and reference modes do not mix. |
| Same as Wan 3.0 when turnaround matters more than cost | `wan3.0-video-prime` | Identical capabilities, faster on longer and higher-resolution jobs, and about 1.5× Wan 3.0's per-second price; prefer plain Wan 3.0 otherwise. |
| Short 720p generation, text rendering, or source-video edit within its narrow live contract | `gemini-omni-flash` | Strong short-form instruction following, multi-shot generation, and conversational editing. |
| H3-family 2K delivery, high-fidelity endpoint control, or focused multimodal references | `hailuo-h3` | Full-quality tier with image/video/audio references; choose it when Max's live contract cannot meet the task. |
| Fast H3-family text-to-video, first/last-frame, or reference-to-video with image/video/audio sources | `hailuo-h3-max` | Faster multimodal tier when its live quality and resolution fit; suitable for source-video variants without discarding the source. |
| Fastest, lowest-cost H3-family text-to-video or first/last-frame generation without generic references | `h3-max-turbo` | Turbo supports text and endpoint frames, not reference image/video/audio inputs; use H3 Max or H3 for references. |
| xAI one-image animation with native sound | `grok-video-1.5` | Newest Grok video generation; the current Renoise contract requires exactly one input image. |
| Upscale an existing video without changing its content | `upscale-video-topaz-starlight-2.5` | Dedicated restoration/upscaling model; use a generation or editing model when objects, motion, timing, or style should change. |

### Tier and version notes

- Within the H3 family, use full H3 when its higher-quality output or 2K matters; choose H3 Max when speed/cost matters and its live image/video/audio reference roles, combinations, and output quality fit. Turbo is the text/first/last-frame budget tier, not a generic reference mode. A source video needs `reference_video`; `first_frame` alone is not equivalent.
- Grok's public upstream capabilities may move faster than Renoise's contract. Route from live roles and limits rather than assuming an upstream feature is connected.
- Model preference leaderboards are task-, resolution-, and audio-filter-specific; use them as volatile evidence, not a single aggregate ranking.

## Video Prompting Styles

### Seedance 2.5 — multimodal director brief

Assign each reference an explicit job, then write shots:

```text
References:
- @material:<ID>: subject identity
- @material:<ID>: location/style
- @material:<ID>: motion/camera/voice

Shot 1: framing, subject action, camera move, environment, sound/dialogue.
Shot 2: ...
Continuity: traits and objects that must remain unchanged.
Constraints: no subtitles/logo/watermark unless requested.
```

- Use a few stable identity features rather than an exhaustive biography.
- Give each beat a clear camera behavior; avoid simultaneous conflicting moves.
- Describe body part, speed, force, and physical consequence for actions.
- Externalize emotion through visible behavior and direct dialogue/audio explicitly.
- Begin with the intended result, map every reference to its purpose, then use non-overlapping timestamp ranges or numbered shots with continuity notes.
- For a source-video edit, name the source, the intended change, its time range when relevant, and what remains unchanged.
- For extension, state forward or backward direction and describe the visual, motion, and audio continuity across the source boundary.

### Wan 3.0 — single-shot brief

```text
[subject and action], [setting], [one camera move], [lighting/style].
References: @material:<ID> identity / @material:<ID> motion / @material:<ID> sound.
Audio: dialogue, effects, or ambience — or no audio.
```

- Keep each job to one brief shot or a short causal sequence; route multi-beat narratives, edits, and extensions to Seedance 2.5.
- Write `@material:<ID>` mentions; the live contract rewrites them into Wan's numbered image/video/audio labels.
- Choose frame mode (first frame, optionally with a last frame) or reference mode, never both; a last frame alone is rejected.
- Never ask Wan to edit, extend, continue, remove, or replace content in an existing video.

### Gemini Omni Flash — conversational generation/editing

- For creation: write a direct natural-language brief with subject, action, camera, setting, visible text, and desired audio.
- Gemini readily creates multi-shot clips; say “single continuous shot,” “single unbroken scene,” or “no scene cuts” when continuity is required.
- Use natural time phrases or `[0–3s]` blocks for important beats.
- For editing: identify the source, say **change only X**, and list what remains unchanged.
- Make one edit per turn when possible; use follow-up refinement rather than replacing the entire prompt.

### MiniMax H3 family — mode-specific audiovisual plan

Choose one mode that the selected model exposes. Full H3 and H3 Max can use generic reference mode when their live roles allow the specific image/video/audio mix; H3 Max Turbo supports text and first/last-frame modes, not generic references. Frame and reference roles are separate modes — do not combine them unless the selected model's live guidance explicitly permits it:

- **Text-to-video:** state the subject, action over time, camera movement, setting, lighting, and sound intent directly.

- **First frame:** describe only the motion, camera path, action development, and sound after the supplied opening state.
- **Last frame:** describe the plausible path that converges on the supplied ending.
- **First + last:** describe the continuous transition between states; avoid re-describing the two stills. Prefer one coherent shot unless a cut is essential.
- **Reference mode (H3 / H3 Max when live):** assign each image, video, and audio an explicit identity, motion, style, voice, action, or sound job. Attach the authorized source video directly when its motion/timing matters, and check the selected model's live duration and role-combination limits.

For complex reference work use:

```text
References: each @material:<ID> image/video/audio and its job.
Task summary: target result and what each source contributes.
Preserve: identity, motion, voice, style, or content that carries over.
[Shot 1] Visual action, camera motion, dialogue, and diegetic sound.
[Shot 2 at time] Cut and continuation.
Overall soundscape: ambience, Foley, and non-verbal sound.
Non-diegetic music: instrumentation, tempo, dynamics — or none.
```

Specify camera **type + amplitude + speed** only when they matter. Keep dialogue exact, label speakers consistently, and separate in-scene sound from background score.

### Grok Imagine Video — animate the change

For image-to-video, do not re-describe the static source:

```text
[subject motion/change], [one camera move], [atmosphere/lighting change].
Sound: specific dialogue/SFX/ambience; no music if unwanted.
```

- Use one primary action or a short causal sequence and front-load the important motion.
- Attach exactly one input image and describe the change from that starting frame.
- Give the attached image a clear visual or motion job.
- Request dialogue, effects, ambience, or music explicitly when native audio matters.

### Topaz Starlight — no creative prompt

Upscaling is not regeneration. Supply only the source video and settings required by the live capability; do not add a creative prompt or unrelated media inputs. If the user wants content changed, route to a video editing model instead.

# Audio Routing

| Task | Prefer | Why |
|---|---|---|
| Song, instrumental, score cue, jingle, or music bed — anything where music is the deliverable | `lyria-pro` | Composes a complete structured song (verses, chorus, bridge, optional vocals and lyrics) up to about 3 minutes; replaces Lyria 3 Clip. |
| Dialogue, expressive narration, podcast, dubbing/re-voicing, SFX/ambience, or a complete sound scene | `seed-audio-1.0` | Unified speech, dialogue, effects, and ambience generation; live default audio model. |

Neither broadly replaces the other: Lyria composes music; Seed Audio handles speech-led and complete sound scenes.

## Lyria 3 Pro — composer brief

```text
Genre/style: ...
Mood: ...
Instrumentation: ...
Tempo/rhythm: ... BPM
Vocals: instrumental OR voice type, delivery, and language
Lyrics/theme: exact lyrics or subject
Target length: ~N seconds (approximate)
Structure: [Intro] → [Verse] → [Chorus] → [Bridge] → [Outro]
Production: mix character, era, space, dynamics
```

- Explicitly say **instrumental** when vocals are unwanted.
- Name instruments, tempo, vocal range/texture, backing vocals, and language.
- There is no duration parameter: state the target length in the prompt and treat it as approximate; match lyric volume to that length.
- Label sections with `[Intro]`, `[Verse]`, `[Chorus]`, `[Bridge]`, and `[Outro]`; put exact lyrics under the section they belong to.
- Use timestamps only for meaningful structural changes; musical alignment follows bars rather than sample-accurate timing.
- Describe genre, era, instrumentation, and production traits instead of requesting a named artist imitation.
- Use an optional guide image only when the live material roles expose it.

## Seed Audio 1.0 — sound-scene script

```text
Setting/acoustics: ...
Continuous sound bed: ...
Speaker A (age, accent, timbre, emotion, pace): "Exact line."
Action-tied SFX: ...
Speaker B (...): "Exact line."
Music cue and ending: ...
```

- For full-scene work, direct it as a scene; for speech-only work, specify voice identity, performance, pacing, and acoustics.
- Label speakers and exact lines; match prompt language to dialogue language.
- Use clean authorized voice references, or an authorized character image for inferred voice, according to the mutually exclusive live reference modes.
- Use concrete Foley and ambience; onomatopoeia can clarify transient sounds.
- Use precise timing for dialogue; describe SFX, ambience, and music timing approximately unless live guidance says otherwise.

# Research Basis

Reviewed 2026-10-13 against live Renoise model contracts. Rankings are directional and decay quickly; live capabilities and task-specific tests override them.

Primary guidance:

- OpenAI GPT Image: https://openai.com/index/introducing-chatgpt-images-2-5/, https://developers.openai.com/api/docs/models/gpt-image-2.5-flare, https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst, https://developers.openai.com/api/docs/guides/image-generation, and https://developers.openai.com/cookbook/examples/multimodal/image-gen-models-prompting-guide
- Google Gemini image: https://ai.google.dev/gemini-api/docs/image-generation and https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-nano-banana
- Midjourney prompting/version docs: https://docs.midjourney.com/hc/en-us/articles/32023408776205-Prompt-Basics, https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version, and https://updates.midjourney.com/version-8-2/
- ByteDance Seedream: https://seed.bytedance.com/en/blog/deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite and https://seed.bytedance.com/en/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro
- xAI Imagine image/video: https://docs.x.ai/developers/model-capabilities/imagine and https://docs.x.ai/developers/model-capabilities/video/generation
- Wan 3.0 / Wan 3.0 Prime: live Renoise model guidance and per-second credit pricing (2026-10-13).
- BytePlus Seedance 2.5 and 2.0 prompt guides: https://docs.byteplus.com/en/docs/ModelArk/2607689 and https://docs.byteplus.com/en/docs/ModelArk/2222480
- MiniMax H3 family: https://www.minimax.io/blog/minimax-h3, https://platform.minimax.io/docs/guides/video-generation, https://huggingface.co/MiniMaxAI/MiniMax-H3/tree/main/docs, and https://blog.fal.ai/introducing-h3-max-by-fal/
- Topaz Starlight Precise 2.5: https://developer.topazlabs.com/video-models/starlight/starlight-precise-2.5
- Gemini Omni: https://ai.google.dev/gemini-api/docs/omni
- Lyria: https://ai.google.dev/gemini-api/docs/music-generation, https://deepmind.google/models/lyria/prompt-guide/, and https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-lyria-3-pro
- Seed Audio: https://seed.bytedance.com/en/blog/from-speech-to-audio-creation-introducing-the-seed-audio-1-0-audio-creation-model and https://docs.byteplus.com/en/docs/byteplusvoice/seedaudio-01
- Independent preference evidence: https://artificialanalysis.ai/image/leaderboard/text-to-image, https://artificialanalysis.ai/image/leaderboard/editing, https://artificialanalysis.ai/video/leaderboard/text-to-video, https://artificialanalysis.ai/video/leaderboard/image-to-video, and https://artificialanalysis.ai/video/leaderboard/video-editing

View saved version →

---
name: storyboard-sheet
description: >
  Runtime-neutral conversion of novels, scripts, outlines, or episode beats into
  short-drama storyboard sheets, shot lists, image prompts, and video-generation
  plans. Supports episode-level review sheets and shot-by-shot planning.
metadata:
  author: renoise
  version: 0.4.1
  category: video-production
  tags: [storyboard, short-drama, adaptation, image-generation, video-generation, portable]
---

# Storyboard Sheet / Short Drama Board

Turn source text into usable visual planning for short drama and AI video generation. Do not paste raw novel/script text into image prompts; first adapt it into visual beats.

Use only capabilities exposed by the current host. Planning works without generation capability. If the user asks to generate outputs, query live capabilities and hand the complete plan to the active host execution workflow. If generation is unavailable, return the complete plan and prompts in the conversation; never guess commands or require host filesystem access.

```text
source → adaptation bible → episode beats → shot list/sheet plan → references → prompts → optional generation → QA → targeted rerun
```

## 1. First Ask: What Output?

If unclear, ask once:

```text
你要哪种输出？
A. 每集一张 storyboard sheet，用于预览/审稿
B. 每个 shot 单独一张图，用于后续视频生成
C. 两者都要：先 sheet 审稿，再拆 shot 图和视频 prompt
```

Default for short-drama/video work: B or C, not only a sheet.

### A. Storyboard Sheet / 审稿预览

```text
one episode = one image containing multiple panels
```

Good for pitch, visual overview, client review, and director preview. It is not ideal as direct video-model input.

Default:

```text
panels: 6
layout: 2×3
ratio: 16:9 unless user wants vertical short drama
text: only tiny P1-P6 labels
```

### B. Shot-by-shot / 视频生成

```text
one shot = one image prompt + one video prompt
```

Good for video generation, first-frame workflows, and precise continuity.

Default for short drama:

```text
ratio: 9:16
shots per 60-90s episode: 8-12
style: vertical mobile drama, close-up heavy, emotion readable
```

## 2. Short Drama Adaptation

Build an adaptation bible in the conversation:

```text
- title / genre
- main characters and relationships
- protagonist desire
- core conflict engine
- must-keep plot points
- secrets / reveals / reversals
- recurring locations
- visual motifs
- continuity constraints: age, wardrobe, injuries, props, money, rules
```

Each episode should usually follow:

```text
opening hook → setup → conflict → escalation → reversal/reveal → cliffhanger
```

Rules:

- Internal thought → visible action, prop, expression, or dialogue.
- Long exposition → conflict scene or reaction beat.
- One beat must be visually clear in one shot/panel.
- Keep the first three seconds strong.
- End with a hook, reversal, danger, or emotional question.
- Prefer faces, reactions, hands, phones, doors, documents, money, wounds, and reveals over abstract narration.

## 3. Character and Scene References

A character appearing more than once needs one approved visual anchor; do not rely on text description alone. Use a user-authorized image or create a reference through the host's generation flow, show it for approval, and reuse the same resulting material through a role advertised by the selected model.

Character reference prompt:

```text
角色设定图，干净背景，现实主义竖屏短剧风格。
[角色名]，[年龄]，[身份]，[体型]，[发型]，[五官/气质]。
固定服装：[服装]。
包含正面全身、半身近景、侧面小头像，必须像同一个人。
不要水印，不要多余文字，不要夸张动漫风。
```

For recurring locations, create a scene reference only when needed. Too many references can confuse layout and identity.

Before optional generation, read `../model-routing/SKILL.md`, then:

1. Query live image-model capabilities.
2. Preserve a user-selected model; otherwise choose the best available specialist for the shot/sheet task and use the advertised default only as fallback.
3. Apply that model's prompting profile and use only advertised ratio, resolution, duration, material roles, and limits.
4. Hand the complete prompt, references, and parameters to the active host execution workflow.
5. Record any task identifier returned by the host before waiting or polling.

## 4. Output Templates

### Episode Sheet Plan

```json
{
  "episode": 1,
  "title": "...",
  "hook": "...",
  "panels": [
    {"panel": "P1", "beat": "opening hook", "visual": "..."},
    {"panel": "P2", "beat": "setup", "visual": "..."},
    {"panel": "P3", "beat": "conflict", "visual": "..."},
    {"panel": "P4", "beat": "escalation", "visual": "..."},
    {"panel": "P5", "beat": "reveal/reversal", "visual": "..."},
    {"panel": "P6", "beat": "cliffhanger", "visual": "..."}
  ]
}
```

Sheet prompt:

```text
一张完整的电影分镜 storyboard sheet，严格 2行3列，共6个清晰 panel，黑色细边框。
现实主义短剧风格，角色与参考图一致，服装/年龄/发型不变。
不要水印，不要长文字，只允许很小的 P1-P6 编号。

第[N]集《[标题]》：
P1 ...
P2 ...
P3 ...
P4 ...
P5 ...
P6 ...
```

### Shot List for Video Generation

```json
{
  "episode": 1,
  "title": "...",
  "shots": [
    {
      "shot": "EP01_SH01",
      "duration": "4s",
      "framing": "tight close-up / medium / over-shoulder / insert",
      "visual": "what we see",
      "action": "one clear action only",
      "emotion": "specific readable emotion",
      "camera": "simple camera move or static",
      "image_prompt": "first-frame image prompt",
      "video_prompt": "motion prompt for video model"
    }
  ]
}
```

Video-friendly rules:

- One shot = one action.
- Avoid time jumps inside one shot.
- Avoid complex crowd/blocking unless necessary.
- Use simple camera language: static, slow push-in, handheld slight shake, over-shoulder.
- Keep text/UI minimal; exact generated text often fails.
- Generate first frames individually, not from a multi-panel sheet.

## 5. QA

Story QA:

```text
[ ] hook is strong enough
[ ] episode order is correct
[ ] no false plot events
[ ] reveal/cliffhanger matches adaptation
[ ] internal monologue became visible action
```

Visual QA:

```text
[ ] correct characters, age, wardrobe
[ ] consistent face/hair/body
[ ] correct location/time/props
[ ] layout correct if sheet
[ ] shot is usable as first frame if video workflow
[ ] no harmful/confusing generated text
```

Give a direct result:

```text
Keep: EP01, EP02
Rerun: EP03_SH04 - character age drifted; EP04 sheet - layout became poster
```

Rerun only failed sheets/shots unless the whole visual system is wrong. Preserve approved references and add the specific failed constraint to the revised prompt.

## 6. User-Facing Summary

Keep the explanation short:

```text
我会先把原文改成短剧节奏：开场钩子、冲突升级、反转和集尾悬念。
如果是审稿，我做每集一张 storyboard sheet；如果要生成视频，我会拆成逐 shot 的首帧图 prompt 和视频 prompt，并用角色 reference 保持一致。
```

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: video-fission
description: >
  Create controlled variants from one owner-authorized source video. Use when
  the user explicitly asks for video fission, multiple video variants,
  controlled variations, or several hypotheses from the same clip. Analyze the
  source first, then resolve what should vary before proposing any generation.
metadata:
  author: renoise
  version: 0.2.2
  category: video-production
  tags: [video, fission, variants, experimentation, portable]
---

# Video Fission

Create a controlled batch of variants from one source video. This is an experiment, not a multi-shot production workflow: vary one axis, or at most two, and lock everything else.

## Runtime Boundary

Use only capabilities exposed by the current host. The source must be one conversation attachment, material, task result, or other video that the host identifies as owner-authorized. Never invent a source, request a host filesystem path, or use more than one source video.

Live model capabilities are authoritative for availability, roles, role combinations, limits, duration, ratio, resolution, and audio controls. Inspect them for every fission run; model names below are routing preferences, not permission to assume a capability.

## Hard Gates

1. **Analyze before resolving the direction.** Once the source is available, use the host's video-analysis capability before finalizing what to vary. If analysis is unavailable or fails, stop. Analysis creates no generation task.
2. **Clarify only when blocked.** If the initial request already names a usable, safe experimental axis, use it after analysis without restating it for text confirmation. Ask one focused question only when the axis is absent, unsafe, or ambiguous; allow a second axis only when the user explicitly needs it.
3. **Default the count.** Use **4 outputs** when no count is supplied; do not ask merely to confirm that default. Four is not a hard cap: accept another reasonable count of at least two, constrained only by live service limits, the total estimate, and the user's budget.
4. **Use one logical batch.** Hand the complete disclosed experiment to the active host execution workflow. Use a native batch primitive when the host exposes one; otherwise follow its documented ordered execution path. Do not invent tool names or execution interfaces.

## 1. Analyze the Source

Report the analysis in the user's language, including:

- subject, setting, actions, camera, composition, lighting, style, pacing/edit structure, duration, aspect ratio, and audio/dialogue;
- dimensions that appear fixed and plausible dimensions to vary;
- uncertain or inferred details as warnings.

Do not resolve an unstated direction until this report exists. If the request already supplies a usable axis, disclose how the analysis maps to it and proceed. Otherwise ask one focused question:

> Which direction should the variants explore? I recommend one axis from this clip's analysis; we can use at most two. The default is 4 outputs, or name another reasonable count.

Offer only source-relevant directions, such as action/motion, camera treatment, pacing, atmosphere/lighting, performance, or transformation. Do not preselect an absent direction for the user.

## 2. Use the Source Video Directly

This workflow has one generation path:

1. Select MiniMax H3 Max (`hailuo-h3-max`) only if the live capability advertises `reference_video`.
2. Use the exact owner-authorized source analyzed in Step 1 as every variant's sole `reference_video` input. If the source reference changes, analyze the replacement before continuing.
3. Refer to that source as `Video 1` in every generation prompt.

Do not extract or upload a frame, and do not replace the source with a `first_frame`. Before handing off paid execution, compare the analyzed source with the live capability's reference-video duration and count limits. If the source exceeds a duration limit, report its measured duration and the live limit when available, ask the user to shorten it below that limit with an editor available to them, then require the shortened source to be referenced and analyzed again. If H3 Max reference-video generation is not live or another source limit is violated, stop and explain the incompatibility. Do not assume a particular host editor, invent another mode, or silently route to another model.

## 3. Design the Experiment

Before submission, show a compact variant matrix in the user's language:

| Variant | Axis value(s) | Distinct hypothesis | Locked dimensions |
|---|---|---|---|
| V1 | ... | If ..., then ... because ... | ... |

Rules:

- Use one axis by default and never more than two.
- Give every variant a genuinely distinct, testable hypothesis; do not use cosmetic synonyms.
- Lock all unselected dimensions across every variant, including model, source video, subject identity, composition, duration, ratio, resolution, audio mode, dialogue, and output controls unless one is the selected axis.
- Keep the same prompt skeleton and change only the clauses that implement the selected axis values.
- If two axes are used, choose intentional combinations; do not create an unrequested Cartesian product.
- Derive every parameter and material role from the selected model's live capability.

## Dialogue and Language

Use the user's language for analysis, questions, matrices, plan text, and results. Prompts are English by default only when they contain no speech.

If any variant contains dialogue, voiceover, or narration, use the explicitly requested or safely inferable spoken language; clarify it only when ambiguous. Keep every spoken line verbatim in that language; never translate it. For dialogue-dense variants, keep the whole prompt in the spoken language. Unless dialogue is the selected axis, lock its text, speaker, delivery, and spoken language across all variants.

## 4. Hand Off One Logical Batch

After the direction, count, hypotheses, prompts, and any spoken language are resolved, hand the complete variant plan to the active host execution workflow. Pass the source aspect ratio, or the closest ratio supported by H3 Max. Preserve any source provenance handle exposed by the host so outputs remain linked to their source where supported.

The execution handoff must:

- keep the whole experiment reviewable as one disclosed plan;
- provide the actual parameters needed for per-variant and total estimation;
- submit the source video directly as every task's `reference_video`;
- follow the host's approval, concurrency, idempotency, and task-tracking rules;
- preserve variant order and never repeat paid execution after interruption.

For revisions, ask which axis value or failed variant to change only when the request is ambiguous, preserve accepted outputs, and hand only the revised experiment back to the host.

View saved version →

---
name: video-remake
description: >
  Recreate or remix one supplied reference video with replacement characters,
  products, branding, props, or scenes. Use when the user asks to replicate a
  video, remake a clip, 复刻视频, 剪同款, 换脸, or make a version of a supplied
  video with controlled replacements. The source video must remain attached.
metadata:
  author: renoise
  version: 0.1.3
  category: video-production
  tags: [video, remake, replication, replacement, portable]
---

# Reference Video Remake / 复刻视频 / 剪同款

This workflow resolves replacement slots before final generation. The current host owns authorized media analysis, materials, live capabilities, costs, spending approvals, and tasks. This skill owns slot matching, creative decisions, and resumable plan state kept in the conversation.

## Runtime Boundary

Use only capabilities and references exposed by the current host. Do not emulate missing analysis/material/task operations. When you have a shell (the host's own or a cloud sandbox tool such as `sandbox_exec`), your own code (editing, compositing, motion graphics, music) is available alongside generation, as in director's Runtime Boundary; use it wherever it brings the remake closer to the reference. Before selecting a model or writing prompts, read `../model-routing/SKILL.md` and inspect live capabilities.

Use the user's language for analysis, plans, approvals, and results. If retained or replacement content contains dialogue, voiceover, or narration, use the explicitly requested or safely inferable spoken language; clarify only when ambiguous, and keep every spoken line verbatim in that language.

## Non-Negotiable Source Policy

The source video is always attached to final generation through a compatible video role advertised by the selected model.

Before planning replacements:

1. Query live video-model capabilities.
2. Select a model that advertises both a compatible source/reference-video role and every image role needed by replacement slots.
3. Check guidance, reference counts, duration, ratio, audio behavior, and role combinations.
4. If no model supports the complete source-video + replacement-image combination, stop and explain. Never omit the source video or invent a role.

The final prompt must state that the source video supplies timing, camera motion, composition, transitions, effects, pacing, and emotional beats. Replacement materials supply identity, product, wardrobe, props, and other slotted content. This text reduces accidental copying but is not a security boundary; normal moderation rules still apply.

## Phase 1 — Analyze and Build a Remake Plan

Use the host's media-analysis capability in template/video mode on the authorized source. If that capability is unavailable, stop; do not request a local command or fall back to copied model code.

That analysis watches and listens, so it carries content, dialogue, music and mood. When you also have a shell, inspect the source yourself to complement it: measure exact cut points, beats, durations and loudness, and look closely at frames of transitions, typography and grading. Use both; the analysis remains the required first step.

The result should contain:

- factual analysis: summary, timeline, composition, style, subjects, and audio;
- a model-neutral prompt with `{{slot_id}}` placeholders;
- slots: each replaceable character/product/object/scene/style plus a reference-image prompt;
- warnings: inferred or uncertain details.

Maintain this plan in the conversation so retries resume known task/material IDs instead of recreating charges:

```json
{
  "version": "v1",
  "source": {
    "authorizedRef": "host reference",
    "materialId": null,
    "role": null
  },
  "analysis": "structured analysis result",
  "videoModel": null,
  "imageModel": null,
  "prompt": "template containing {{slot_id}} placeholders",
  "slots": [
    {
      "id": "subject_1",
      "type": "character",
      "description": "replacement lead",
      "prompt": "English reference-image prompt",
      "source": "user|library|generated|missing",
      "materialId": null,
      "materialName": null,
      "role": null,
      "taskId": null
    }
  ],
  "videoTaskId": null
}
```

Do not store API keys, signed URLs, credential contents, or arbitrary host paths.

## Phase 2 — Fill Replacement Slots

Use this priority:

1. User-provided host-authorized material that clearly matches the slot.
2. Existing owner-scoped Renoise material confirmed by the user.
3. Generate a replacement image from the slot prompt through the active host execution workflow.

Never let filename similarity alone choose a character or product. Show the proposed mapping with source, description, and intended role.

### Slot Plan and Cost

Present in the user's language:

- source-video summary and warnings;
- extracted timeline/style at useful density;
- every replacement slot and proposed source;
- selected image/video models from live capabilities;
- live estimated cost for missing slot images;
- the fact that the source video will be attached to final generation.

If a proposed slot mapping is ambiguous, ask the user to resolve that mapping. Otherwise, hand each requested missing slot and its complete parameters to the active host execution workflow.

For every generated slot:

1. Follow the host's execution, estimate, and approval policy.
2. Record any task identifier returned by the host immediately.
3. Resume that task after a wait interruption; never repeat paid execution blindly.
4. Register the approved result as a reusable material only once.

For user-provided references, register the authorized item once. Record the returned material ID and full server filename. Prompt mentions must use that complete filename, including extension.

Show generated slot images and let the user replace/regenerate individual slots. Do not proceed until all required slots are approved.

## Phase 3 — Prepare Final Generation

Inspect the selected video model again immediately before generation. Then:

1. Register the source video once.
2. Assign it a compatible advertised source/reference-video role.
3. Assign every approved slot a compatible advertised image role.
4. Verify combined reference counts and role combinations.
5. Replace every `{{slot_id}}` with the corresponding exact `@material:{id}` token returned for its registered replacement material.
6. Add an explicit source lock: source video controls motion/editing/style; slot materials control identities and replaceable content.
7. Preserve source dialogue only when the user explicitly wants it and spoken language has been confirmed.

The final prompt must contain no unresolved placeholder.

Request a live estimate with actual duration, resolution, output controls, and full material set; also read the live balance.

### Final Prompt, Assets, and Video Cost

Present in the user's language:

- full final prompt;
- source video → exact material ID and role;
- every slot → exact filename, material ID, and role;
- generation parameters;
- final estimated cost and account balance.

Hand these complete arguments to the active host execution workflow; do not assume a particular tool, command, card, or approval interface.

## Phase 4 — Generate and Resume Safely

Execute through the active host workflow and record any returned task identifier immediately. Resume that same task after interruption or timeout; never repeat paid execution blindly.

Return the result URL, task ID, source/slot mapping, and warnings. For revisions, keep approved slot materials and change only the identified prompt dimension unless the user asks to replace an asset.

## Failure Rules

Do not pre-screen the source, replacement materials, or prompts. Continue normally unless the host returns an explicit content-review error.

| Failure | Action |
|---|---|
| Media analysis unavailable | Stop and explain; do not fall back to a command or copied model code. |
| No model supports source video + slot images | Stop; do not omit the source or invent roles. |
| Source video exceeds supported limits | Ask for/approve a shorter source; do not silently clamp it. |
| Slot generation fails | Keep successful slots and retry only the failed slot after approval. |
| Wait times out | Resume the recorded task ID. |
| Host returns `INPUT_*` / `OUTPUT_*` content-review error | Report the actual error and stop; do not retry paid create or suggest bypasses. |
| Output copies source identity despite slots | Strengthen the source lock or replace the conflicting source; never claim text guarantees isolation. |