version: "1.0.0" name: seed2-video description: Use this skill when a user wants to generate videos with Seed2.io / Super Seed 2 Lite, integrate the Seed2 REST API, choose Seedance 2.0 model tiers, write prompts with image/video/audio references, or troubleshoot Seed2 video generation tasks. metadata: short-description: Generate Seed2.io AI videos website: https://seed2.io tags:

• seed2
• seedance
• video-generation
• text-to-video
• image-to-video
• codex-skill
• claude-code-skill

Seed2 Video

Use the Seed2.io REST API for text-to-video, image-to-video, video reference, audio reference, and first/last-frame transitions.

API Basics

• Base URL: https://api.seed2.io
• Create video: POST /v1/video/generate
• Query task: GET /v1/tasks/:task_id
• Auth header: Authorization: Bearer $SEEDANCE_API_KEY
• Backend model: st-ai/super-seed2-lite
• Duration: integer 4 to 15
• Ratio: 16:9, 9:16, 1:1, 4:3, 3:4, 21:9

Model Selection

Prefer Seedance_2.0_mini_lite for most general use. It is fast, cost-effective, fixed at 720p, and usually finishes in 2-8 minutes.

Use these tiers:

Legacy aliases seedance_2.0_fast and seedance_2.0 are accepted, but prefer the newer IDs above.

Pricing

Assumptions: 1 USD = 10,000 credits, approximate USD/CNY = 6.77.

Prompt References

The API accepts English and Chinese aliases, then normalizes upstream to Chinese bracket references:

• @image_file_1, @image1, @图片1, @图片一 -> 【@图片1】
• @video_file_1, @video1, @视频1 -> 【@视频1】
• @audio_file_1, @audio1, @音频1 -> 【@音频1】

When writing prompts for users, English aliases are fine for API examples, but Chinese bracket references are the upstream canonical form.

Request Patterns

Text to video:

curl -X POST https://api.seed2.io/v1/video/generate \
  -H "Authorization: Bearer $SEEDANCE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A cat playing piano in a jazz bar, cinematic lighting",
    "model": "Seedance_2.0_mini_lite",
    "resolution": "720p",
    "duration": 5,
    "ratio": "16:9"
  }'

Image/video/audio references:

{
  "prompt": "The person in @image_file_1 follows the motion of @video_file_1 and speaks with @audio_file_1",
  "model": "seedance2.0_fast_vision",
  "resolution": "720p",
  "duration": 8,
  "ratio": "16:9",
  "image_files": ["https://example.com/person.jpg"],
  "video_files": ["https://example.com/motion.mp4"],
  "audio_files": ["https://example.com/speech.mp3"]
}

First/last frames:

{
  "prompt": "Smooth cinematic transition from first frame to final frame",
  "model": "seedance2.0_vision",
  "resolution": "1080p",
  "functionMode": "first_last_frame",
  "duration": 5,
  "ratio": "16:9",
  "image_url": "https://example.com/first.png",
  "end_image_url": "https://example.com/last.png"
}

Workflow

1. Ask for missing essentials only: prompt, media URLs, target ratio, duration, desired quality/speed.
2. Pick a model tier and resolution from the tables above.
3. Submit POST /v1/video/generate.
4. Poll GET /v1/tasks/:task_id until completed or failed.
5. Return result.video_url, points used, and the request payload if useful.

Constraints

• Public media URLs must be accessible by the API.
• image_files max 9, video_files max 3, audio_files max 3.
• Audio cannot be used alone; include image or video references.
• Mini/Fast/Direct tiers are fixed at 720p.
• seedance2.0_fast_vision supports only 480p and 720p.
• seedance2.0_vision supports 480p, 720p, and 1080p.