Doubao Native Media Skill

This is a native OpenClaw skill. Do not spin up the upstream MCP server unless the user explicitly asks for MCP compatibility.

Use this skill for

• Doubao / 豆包 text-to-image
• image-to-image or multi-reference image generation
• Doubao text-to-video or image-to-video
• querying an async Doubao video task by task_id
• troubleshooting Volcengine Ark endpoint/model issues

Commands

Generate an image

python3 {baseDir}/scripts/doubao_media.py image \
  --prompt "A cinematic cyberpunk alley in rain" \
  --size 2560x1440

Generate a video

python3 {baseDir}/scripts/doubao_media.py video \
  --prompt "A panda astronaut waves on the moon" \
  --video-duration 5 \
  --fps 24 \
  --resolution 1080p

Query a video task

python3 {baseDir}/scripts/doubao_media.py task --task-id your-task-id

Wait for a video task and optionally download the result

python3 {baseDir}/scripts/doubao_media.py wait \
  --task-id your-task-id \
  --timeout 600 \
  --interval 5 \
  --download-to ./doubao-result.mp4

Input rules

• Always prefer --endpoint-id when the user has a provisioned Volcengine Ark endpoint.
• Fall back to model names only when endpoint ids are unavailable.
• For video generation, this skill mirrors the upstream behavior and appends --dur, --fps, --rs, and --ratio to the prompt when they are not already present.
• If the user supplies image URLs, pass them through exactly; do not download or re-host unless asked.

Troubleshooting

• If neither --endpoint-id nor a default endpoint env var exists, the script falls back to the default model env var.
• If the API returns InvalidEndpointOrModel.NotFound, ask the user to verify the Volcengine Ark endpoint authorization first.
• Video generation is async. If generation succeeds, capture task_id and query it later with the task subcommand, or use wait for automatic polling.

References

• Read references/api-notes.md when you need request shapes, defaults, or caveats.