@renoise/video-maker 0.1.0

package/skills/renoise-gen/SKILL.md

@@ -0,0 +1,362 @@
+---
+name: renoise-gen
+description: Generate AI videos and images via Renoise platform. Create tasks, upload materials, browse characters, poll results, download outputs. Supports text-to-video, image-to-video, video-to-video, and text-to-image. Use this skill whenever the user asks to "generate video", "create video", "text to video", "image to video", "generate image", "AI video", "AI image", or describes any video/image content they want generated with AI.
+allowed-tools: Bash, Read, Write, Glob
+metadata:
+  author: renoise
+  version: 0.1.0
+  category: video-production
+  tags: [general, video-generation]
+---
+
+# AI Video & Image Generation (via Renoise)
+
+Generate AI videos and images through the Renoise platform.
+
+## Supported Models
+
+| Model | Type | Description |
+|-------|------|-------------|
+| `renoise-2.0` | Video | Default video model with storyboard control |
+| `nano-banana-2` | Image | AI image generation model |
+
+## Core Concept: Two Video Modes — Finished Cut vs Clip Stock
+
+Choose between two generation modes based on user needs:
+
+### Finished Cut Mode — Default
+
+For producing a complete, ready-to-use video. Leverages powerful storyboard control to **direct content, camera movement, and pacing across different time segments within a single 15s clip**:
+
+- Music/SFX flow naturally with coherent progression
+- Character consistency maintained within the same segment
+- Complex continuous camera movements (e.g., close-up → orbit → wide pull back)
+- Only 1 API call needed
+
+**Default 15s, use time-annotated prompt to control content.**
+
+```
+[0-3s] Close-up of hands unboxing a sleek black device on a white desk. Camera snaps dolly in to reveal the logo.
+
+[3-10s] The woman picks it up, examines it from different angles. Medium shot, smooth orbit around the product. Spoken dialogue: "I've been waiting for this." Mouth clearly visible, lip-sync aligned.
+
+[10-15s] She places it on a wireless charger, LED glows blue. Pull back to wide shot of the full workspace. The frame holds steady.
+```
+
+### Clip Stock Mode
+
+For producing atomic clips for post-production editing. Each clip focuses on **a single action + single camera move** for maximum flexibility:
+
+- Each clip **3-5s**, one clip does one thing
+- No time annotations needed, just describe a single scene
+- Batch generate multiple clips, organize with tags
+- Combine freely in post-production
+
+```bash
+# Clip stock example: prepare clips for a product video
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create --prompt "Extreme close-up of a matte black smartwatch on white marble, slow dolly in, studio lighting." --duration 5 --tags product-x,detail
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create --prompt "A hand picks up the smartwatch from the table, medium shot, tracking follows the hand upward." --duration 5 --tags product-x,pickup
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create --prompt "Wrist-level shot of the watch on a person's arm, smooth orbit, outdoor golden hour." --duration 5 --tags product-x,lifestyle
+```
+
+### How to Decide?
+
+| Signal | Mode |
+|--------|------|
+| "generate a video", "make a short film" | Finished Cut — 15s storyboard |
+| "prepare clips", "clip", "for editing", "B-roll" | Clip Stock — 3-5s atomic clips |
+| "shot list", "storyboard" | Clip Stock — generate per shot |
+| Unclear | **Default to Finished Cut**, confirm with user |
+
+## Model Specs
+
+### renoise-2.0 (Video)
+
+| Parameter | Value |
+|-----------|-------|
+| Duration range | 5-15s, any integer |
+| **Default duration** | **15s** (maximize storyboard capability) |
+| Aspect ratio | `1:1`, `16:9`, `9:16` |
+| Recommended mode | **Image-to-Video with storyboard grid** (best visual consistency) |
+| Prompt language | **English**, natural narrative paragraphs |
+
+> **Default to Image-to-Video with storyboard grid**: Generate reference images as a 9-grid or 16-grid composite, upload as `ref_image`. Grid images bypass privacy detection more reliably than individual face photos because faces are small within the grid cells. **Fallback to Text-to-Video** only when the grid is still blocked by privacy detection (`PrivacyInformation` error) — text-only prompts are never subject to this limitation.
+
+### nano-banana-2 (Image)
+
+| Parameter | Value |
+|-----------|-------|
+| Resolution | `1k`, `2k` |
+| Aspect ratio | `1:1`, `16:9`, `9:16` |
+| Prompt language | **English** |
+
+See `${CLAUDE_SKILL_DIR}/references/video-capabilities.md` for details.
+
+## Configuration
+
+CLI path: `${CLAUDE_SKILL_DIR}/renoise-cli.mjs` (Node.js 18+)
+
+API Key and base URL are configured via environment variables (`RENOISE_API_KEY`, `RENOISE_BASE_URL`).
+
+## CLI Commands
+
+All commands: `node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs <domain> <action> [options]`
+
+Four domains: `task`, `material`, `character`, `credit`
+
+### Check Balance (always check before creating tasks)
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs credit me                              # User info + balance
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs credit estimate --duration 10           # Estimate cost
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs credit history                          # Transaction history
+```
+
+### Video Input Modes (Mutually Exclusive)
+
+Three ways to provide visual input. **These are mutually exclusive — do NOT mix them in the same task.**
+
+| Mode | `--materials` | Description |
+|------|---------------|-------------|
+| **First frame only** | `ID:first_frame` (1 image) | Pin the first frame; prompt is optional |
+| **First + last frame** | `ID1:first_frame,ID2:last_frame` (2 images, both roles required) | Pin start and end frames; model generates the transition |
+| **Multimodal reference** | `ID:ref_image`, `ID:ref_video`, etc. | Reference images (1-9), videos (0-3), audio (0-3) for style/content guidance |
+
+> **Cannot combine**: e.g. you cannot use `first_frame` together with `ref_image` in the same task. If you need "first/last frame + reference style", use multimodal reference mode and describe in the prompt which image should be the first/last frame — but for strict frame accuracy, prefer the dedicated first+last frame mode.
+
+**Image requirements** (for first/last frame and ref_image):
+- Format: jpeg, png, webp, bmp, tiff, gif
+- Aspect ratio (W/H): 0.4 ~ 2.5
+- Dimensions: 300 ~ 6000 px per side
+- Size: < 30 MB per image
+
+### Generate Video (one step: create + wait for completion)
+```bash
+# Text-to-video finished cut (default renoise-2.0, 15s storyboard)
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "[0-5s] Close-up of a cat on the moon, slow push in. [5-12s] The cat starts dancing, smooth orbit camera, stars twinkling. [12-15s] Wide pull back revealing the full lunar landscape, frame holds steady. Cinematic lighting, shallow depth of field." \
+  --duration 15 --ratio 1:1
+
+# First frame only — pin the opening shot
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/start.jpg
+# (note the material ID from output, e.g. #42)
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "The woman walks toward the camera, gentle breeze, cinematic lighting." \
+  --materials "42:first_frame" --duration 10 --ratio 16:9
+
+# First + last frame — pin both start and end
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/start.jpg   # e.g. #42
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/end.jpg     # e.g. #43
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "Smooth transition from dawn to sunset over the city skyline." \
+  --materials "42:first_frame,43:last_frame" --duration 10 --ratio 16:9
+
+# Multimodal reference — image-to-video
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/photo.jpg
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "[0-5s] Close-up of the product on a white surface, gentle dolly in. [5-12s] Camera orbits around the product revealing all angles, soft studio lighting. [12-15s] Pull back to wide shot, product centered, frame holds steady." \
+  --materials "ID:ref_image" --duration 15 --ratio 16:9
+
+# Multimodal reference — video-to-video
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "recreate this motion with a robot character" \
+  --materials "ID:ref_video" --duration 5
+```
+
+### Generate Image
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+  --prompt "A cute cat sitting on a crescent moon, watercolor style, dreamy atmosphere" \
+  --model nano-banana-2 --resolution 2k --ratio 1:1
+```
+
+### Create Task Only (no waiting)
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create \
+  --prompt "[0-5s] ... [5-12s] ... [12-15s] ..." \
+  --duration 15 --ratio 16:9 --tags cinematic
+```
+
+**generate/create parameters**:
+- `--prompt` (required) — English natural narrative prompt; use `[time segment]` annotations for video storyboards
+- `--model` — Model name (default: renoise-2.0; use nano-banana-2 for images)
+- `--duration` — Video duration 5-15s (CLI default: 5). **Always set `--duration 15` for Finished Cut mode**
+- `--ratio` — `1:1` / `16:9` / `9:16` (default: 1:1)
+- `--resolution` — Image resolution `1k` / `2k` (image models only)
+- `--tags` — Comma-separated tags for organization
+- `--materials` — Material references `id:role`, comma-separated for multiple
+- `--characters` — Character references `id1,id2` or `id1:role,id2:role`
+
+### Task Management
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task list                          # List tasks
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task list --status completed       # Filter by status
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task list --tag project-x          # Filter by tag
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task get <id>                      # Task detail
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task result <id>                   # Get result URL
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task wait <id>                     # Poll until complete
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task wait <id> --interval 15 --timeout 300
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task cancel <id>                   # Cancel (pending only)
+```
+
+### Material Upload & Management
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/file.jpg          # Upload (auto-detect type)
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload /path/to/clip.mp4 --type video
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material list                               # List materials
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material list --type image --search cat
+```
+
+### Character Browsing
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs character list                              # List available characters
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs character list --category female --search Jasmine
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs character get <id>                          # Character detail
+```
+
+### Tag Management
+```bash
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task tags                          # List all tags
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task tag <id> --tags a,b,c         # Update task tags
+```
+
+## Task Statuses
+
+`pending` → `assigning` → `assigned` → `queued` → `running` → `completed` / `failed`
+
+Only `pending` tasks can be cancelled (auto-refund).
+
+## Workflows
+
+### Finished Cut Mode (Default)
+```
+credit me → credit estimate → task generate (15s + storyboard prompt)
+```
+
+`task generate` = create + wait + output result, all in one step.
+
+### Clip Stock Mode
+```
+credit me → credit estimate → task create (3-5s × N, grouped by tag) → task wait each → collect task result
+```
+
+Batch generate atomic clips, combine in post-production.
+
+### Image Generation
+```
+credit me → task generate --model nano-banana-2 --prompt "..." --resolution 2k
+```
+
+### With First/Last Frame
+```
+material upload (start.jpg) → material upload (end.jpg) → task generate (--materials "ID1:first_frame,ID2:last_frame")
+```
+
+First frame only: omit the second upload, use `--materials "ID:first_frame"`.
+
+### With Multimodal Reference
+```
+material upload → task generate (--materials "ID:ref_image" or "ID:ref_video")
+```
+**Preferred**: Use storyboard grid images as ref_image — see "Storyboard Grid Workflow" section below.
+
+### With Character Reference
+```
+character list → task generate (--characters "ID")
+```
+
+### Finished Cut > 15s
+
+When target duration exceeds 15s, split into 15s segments, each with its own storyboard prompt:
+
+```bash
+# 30s = 2 × 15s
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create --prompt "[0-5s] ... [5-12s] ... [12-15s] ..." --duration 15 --tags vid-001,s1
+node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task create --prompt "[0-5s] ... [5-12s] ... [12-15s] ..." --duration 15 --tags vid-001,s2
+```
+
+**Maintain consistency**: Repeat full character appearance description at the start of each segment's prompt, use consistent lighting/style keywords, bridge with `Continuing from the previous shot:`.
+
+## Storyboard Grid Workflow (Preferred Approach)
+
+The storyboard grid method produces the best visual consistency across clips by anchoring each generation to a reference image from a unified grid.
+
+### Why Storyboard Grid?
+
+| Approach | Visual Consistency | Privacy Detection Risk | Setup Effort |
+|----------|-------------------|----------------------|--------------|
+| **Storyboard Grid (preferred)** | Highest — all panels share style context | Low — faces are small in grid cells | Medium |
+| Text-to-Video (fallback) | Lower — model interprets style differently each call | None | Low |
+| Individual ref_image | Medium | High — close-up faces trigger blocking | Medium |
+
+### Step-by-Step
+
+1. **Generate reference images as a grid**: Use Midjourney (v7) or Gemini to create a single composite image containing all shots as panels in a 3x3 (9-grid) or 4x4 (16-grid) layout. Each panel shows one key moment from a shot with consistent character appearance and style.
+
+2. **Upload the grid as material**:
+   ```bash
+   node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs material upload storyboard_grid.png
+   # Returns material ID
+   ```
+
+3. **Generate video with ref_image + time-annotated prompt**:
+   ```bash
+   node ${CLAUDE_SKILL_DIR}/renoise-cli.mjs task generate \
+     --prompt "Follow the attached storyboard panels. [0-5s] ... [5-10s] ... [10-15s] ..." \
+     --materials "MATERIAL_ID:ref_image" \
+     --duration 15 --ratio 16:9
+   ```
+
+4. **For videos > 15s**: Split into multiple grids (e.g., 3x3 for shots 1-9, another for shots 10-18), generate each 15s segment with its corresponding grid.
+
+5. **Fallback**: If a grid triggers `PrivacyInformation`, retry without `--materials` (pure text-to-video), copying full character descriptions into the prompt.
+
+### When to Use Each Approach
+
+| Signal | Approach |
+|--------|----------|
+| Any project with recurring characters | **Storyboard grid** (preferred) |
+| Product shots, landscapes, no people | Image-to-Video with individual ref_image |
+| Grid ref_image blocked by privacy detection | **Text-to-Video** (fallback) |
+| Quick one-off generation, no consistency needs | Text-to-Video |
+
+---
+
+## Prompt Writing
+
+**General rules**:
+- **Must be English** — The model understands English narrative paragraphs best. Chinese or tag lists degrade quality and coherence.
+- **Natural narrative paragraphs** — Use complete descriptive sentences, not comma-separated keyword lists (e.g., `woman, café, coffee, 4k`). The model needs to understand causal and temporal relationships.
+- **Specific > Abstract** — `a golden retriever running through shallow ocean waves at sunset` is far better than `a dog on a beach`. More detail = more accurate output.
+- **Structure**: Subject (detailed appearance) + Action (action sequence) + Camera (camera movement) + Scene (environment/lighting) + Style (visual style)
+
+**Finished cut prompt**:
+- Use `[time segment]` annotations for storyboard beats — core advantage, the model switches content and camera per segment
+- Each segment = Subject + Action + Camera
+- At least 2-3 camera changes (e.g., close-up → orbit → wide pull back) for editing rhythm
+- End last segment with `frame holds steady` for easy continuation or natural ending
+
+**Clip stock prompt**:
+- No time annotations needed, describe a single scene directly
+- One clip does one thing: one action + one camera move, for flexible post-editing
+- Keep concise, 3-5 sentences
+
+**Image prompt**:
+- Describe content, style, and atmosphere
+- No time annotations or camera movements needed
+
+See `${CLAUDE_SKILL_DIR}/references/video-capabilities.md` for detailed prompt writing guide and camera movement cheat sheet.
+
+## Error Handling
+
+| Error | Cause | Solution |
+|-------|-------|----------|
+| `PrivacyInformation` | Reference image/video with human faces blocked by privacy detection | Switch to Text-to-Video, describe person's appearance in text |
+| `Insufficient credits` (402) | Balance too low | Inform user of current balance and required cost, suggest top-up |
+| Task `failed` | Generation failed | Use `task get <id>` to check error field. Common causes: prompt violation, server timeout. Adjust prompt and retry |
+| `Auth Error` (401) | Invalid API Key | Check that the `RENOISE_API_KEY` environment variable is set correctly |
+| `wait` timeout | Generation time exceeds timeout | 15s videos typically need 5-10 minutes, increase `--timeout` (e.g., 900) |
+
+## References
+
+- [Video Model Capabilities](references/video-capabilities.md) — Model specs, detailed prompt writing guide, camera movement cheat sheet. Consult when writing prompts.
+- [API Endpoint Reference](reference.md) — Renoise API endpoints and request/response formats. Consult when calling API directly or debugging HTTP errors.

package/skills/renoise-gen/references/api-endpoints.md

@@ -0,0 +1,138 @@
+# Renoise API Reference
+
+## API Endpoints
+
+Base URL: `https://staging--ujgsvru36x4korjj10nq.edgespark.app`
+API Prefix: `/api/public/v1`
+
+All endpoints require `X-API-Key: <api_key>` header.
+
+### Credits
+
+| Method | Path | Description |
+|--------|------|-------------|
+| GET | `/api/public/v1/me` | User info + credit balance |
+| GET | `/api/public/v1/credit/estimate?model=X&duration=Y&hasVideoRef=0` | Cost estimate |
+| GET | `/api/public/v1/credit/history?limit=50&offset=0` | Credit transactions |
+
+### Tasks
+
+| Method | Path | Description |
+|--------|------|-------------|
+| POST | `/api/public/v1/tasks` | Create task |
+| GET | `/api/public/v1/tasks?status=X&tag=Y&limit=50&offset=0` | List tasks |
+| GET | `/api/public/v1/tasks/:id` | Task detail |
+| GET | `/api/public/v1/tasks/:id/result` | Task result (video/image/cover URLs) |
+| POST | `/api/public/v1/tasks/:id/cancel` | Cancel pending task |
+| PATCH | `/api/public/v1/tasks/:id/tags` | Update tags |
+| GET | `/api/public/v1/tags` | List all tags |
+
+### Materials
+
+| Method | Path | Description |
+|--------|------|-------------|
+| POST | `/api/public/v1/materials/upload` | Upload material (multipart) |
+| GET | `/api/public/v1/materials?type=X&search=Y` | List materials with download URLs |
+
+### Characters
+
+| Method | Path | Description |
+|--------|------|-------------|
+| GET | `/api/public/v1/characters?category=X&usage_group=Y&search=Z&page=1&page_size=20` | List characters |
+| GET | `/api/public/v1/characters/:id` | Character detail |
+
+## Request/Response Formats
+
+### POST /api/public/v1/tasks
+
+Request:
+```json
+{
+  "prompt": "string (required)",
+  "model": "renoise-2.0",
+  "duration": 5,
+  "ratio": "1:1",
+  "resolution": "2k",
+  "materials": [
+    { "id": 42, "role": "ref_video" },
+    { "character_id": 3, "role": "reference_image" }
+  ],
+  "tags": ["demo"]
+}
+```
+
+Response (201):
+```json
+{
+  "task": {
+    "id": 1,
+    "prompt": "...",
+    "model": "renoise-2.0",
+    "status": "pending",
+    "estimatedCredit": 5.0,
+    "createdAt": "2026-03-10T..."
+  }
+}
+```
+
+Error (402 — insufficient credits):
+```json
+{
+  "error": "Insufficient credits",
+  "available": 2.5,
+  "required": 5.0
+}
+```
+
+### GET /api/public/v1/tasks/:id/result
+
+Response:
+```json
+{
+  "taskId": 1,
+  "status": "completed",
+  "videoUrl": "https://...",
+  "imageUrl": "https://...",
+  "coverUrl": "https://...",
+  "resolutions": { "720p": "https://..." },
+  "itemCount": 1,
+  "fetchedAt": "2026-03-10T...",
+  "cached": true
+}
+```
+
+## Task Statuses
+
+| Status | Description |
+|--------|-------------|
+| `pending` | Waiting for assignment |
+| `assigning` | Being assigned to account |
+| `assigned` | Assigned, waiting to start |
+| `queued` | Queued on provider |
+| `running` | Generating |
+| `completed` | Done — result available |
+| `failed` | Failed — check error field |
+| `cancelled` | User cancelled |
+
+## Models
+
+- `renoise-2.0` — Default video model (Renoise 2.0)
+- `nano-banana-2` — Image generation model
+
+## Material Roles
+
+- `ref_video` — Reference video (affects pricing)
+- `ref_image` — Reference image
+- `image1`, `image2` — Additional reference images
+- `reference_image` — Character reference image
+
+## Aspect Ratios
+
+- `1:1` (default)
+- `16:9`
+- `9:16`
+
+## Image Resolutions
+
+- `1k`
+- `2k`