IMA Seedance 2.0 Video Generator

# IMA Seedance 2.0 — OpenClaw Calling Protocol

## Hard Rule

OpenClaw must call `scripts/ima_video_create.py`.

OpenClaw must **not**:

- build `/open/v1/tasks/create` payloads
- compute `attribute_id`
- compute `credit`
- build `src_img_url`
- build `src_image`
- build `src_video`
- build `src_audio`

## Required Read Order

1. [`references/protocols/execution.md`](references/protocols/execution.md)
2. [`references/protocols/event-stream.md`](references/protocols/event-stream.md)
3. On demand:
   - [`references/contracts/create-task.md`](references/contracts/create-task.md)
   - [`references/contracts/credit-rules.md`](references/contracts/credit-rules.md)
   - [`references/contracts/payload-rules.md`](references/contracts/payload-rules.md)
   - [`references/flows/reference-image-to-video.md`](references/flows/reference-image-to-video.md)
   - [`references/limits/reference-media-rules.md`](references/limits/reference-media-rules.md)

## Supported User Intents

- text-to-video
- image-to-video
- first-last-frame transition
- reference-image-to-video
- multimodal reference-image-to-video with image / video / audio inputs

## Input Entry Points

- `--prompt`
- `--model-id`
- `--task-type` when explicit task type is required
- `--input-images`
- `--reference-image`
- `--reference-video`
- `--reference-audio`
- `--extra-params`

## Stop Conditions

Stop before task creation if:

- prompt is missing
- model cannot be resolved
- reference media preflight validation fails
- any reference media compliance verification fails
- create-task returns a hard failure

## Task Type Rules

| Input pattern | task_type |
|---|---|
| text only | `text_to_video` |
| one image | `image_to_video` |
| explicit first-last-frame with 2 images | `first_last_frame_to_video` |
| any video input | `reference_image_to_video` |
| any audio input | `reference_image_to_video` |
| multiple images without explicit first-last-frame intent | `reference_image_to_video` |

## Model ID Reference (CRITICAL)

Use **exact model_id** from this table. Do NOT infer from friendly names.


| Friendly Name     | model_id       | Notes                        |
| ----------------- | -------------- | ---------------------------- |
| Seedance 2.0      | `ima-pro`      | ✅ Quality priority, 300~900s |
| Seedance 2.0 Fast | `ima-pro-fast` | ⚠️ Speed priority, 120~600s   |


**User input aliases:**
- Quality/Professional/Pro/专业版/高质量 → `ima-pro`
- Fast/Speed/Quick/极速/快速 → `ima-pro-fast`
- Default/默认 → `ima-pro`

## Model Selection Priority

1. **User preference** (if explicitly stated) → highest priority
2. **Fallback default:** `ima-pro`


| Task | Default | Fast Alternative |
|---|---|---|
| `text_to_video` | `ima-pro` | `ima-pro-fast` |
| `image_to_video` | `ima-pro` | `ima-pro-fast` |
| `first_last_frame_to_video` | `ima-pro` | `ima-pro-fast` |
| `reference_image_to_video` | `ima-pro` | `ima-pro-fast` |

## Minimal Invocation Examples

```bash
# Text to video
python3 {baseDir}/scripts/ima_video_create.py \
  --prompt "a puppy runs across a sunny meadow, cinematic"

# Single image
python3 {baseDir}/scripts/ima_video_create.py \
  --prompt "camera slowly zooms in" \
  --input-images https://example.com/photo.jpg \
  --model-id ima-pro-fast

# Explicit first-last-frame
python3 {baseDir}/scripts/ima_video_create.py \
  --task-type first_last_frame_to_video \
  --prompt "smooth transition" \
  --input-images https://example.com/first.jpg https://example.com/last.jpg

# Multimodal reference mode
python3 {baseDir}/scripts/ima_video_create.py \
  --reference-image https://example.com/product.jpg \
  --reference-video https://example.com/clip.mp4 \
  --reference-audio https://example.com/narration.mp3 \
  --model-id ima-pro-fast
```

## References

- Execution protocol: [`references/protocols/execution.md`](references/protocols/execution.md)
- Event stream: [`references/protocols/event-stream.md`](references/protocols/event-stream.md)
- Create-task contract: [`references/contracts/create-task.md`](references/contracts/create-task.md)
- Credit rules: [`references/contracts/credit-rules.md`](references/contracts/credit-rules.md)
- Payload rules: [`references/contracts/payload-rules.md`](references/contracts/payload-rules.md)
- Reference-image-to-video flow: [`references/flows/reference-image-to-video.md`](references/flows/reference-image-to-video.md)
- Text-to-video flow: [`references/flows/text-to-video.md`](references/flows/text-to-video.md)
- Reference media limits: [`references/limits/reference-media-rules.md`](references/limits/reference-media-rules.md)
- FAQ: [`references/support/faq.md`](references/support/faq.md)
- Troubleshooting: [`references/support/troubleshooting.md`](references/support/troubleshooting.md)