Liblib Comfy Fusion Generation

Problem

liblib-comfy-fusion addresses a common Feishu integration gap: a user provides a local image, a group-chat text excerpt, or a public image URL and expects the skill to run a LiblibAI ComfyUI template workflow, then display the generated image back in the current chat. Without this skill, teams often have to upload local files, construct signed requests, poll task status, and manually download temporary image URLs before sending files through another channel. Those steps can fail when URLs are not public, signatures are malformed, or the chat channel does not render media directly.

How It Works

The skill uses scripts/liblib_client.py to call the LiblibAI ComfyUI App API. The core flow is:
- Input handling: accepts a public URL via --image-url, a local file via --local-image, or the first http(s) URL from --feishu-text / FEISHU_TEXT. Local files are uploaded to R2/S3-compatible storage by default so Liblib can fetch a public URL.
- Async generation: submits a task to POST /api/generate/comfyui/app, authenticates with AccessKey, Signature, Timestamp, and SignatureNonce, then polls POST /api/generate/comfy/status until generateStatus is 5 for success or 6 for failure.
- Result delivery: on success, downloads images[].imageUrl into workspace/outputs/images/YYYY-MM-DD/ and prints MEDIA:... to stdout for Feishu display; the full task JSON goes to stderr for debugging.

Boundaries And Notes

• Liblib must be able to reach image URLs over the public internet; private or internal URLs usually will not work.
• Local-file mode relies on boto3 and R2/S3 upload by default; use data-uri mode or provide a public URL when that stack is unavailable.
• Returned image URLs are temporary, typically valid for about 7 days, so they are not suitable for long-term storage.
• If the API returns content-review errors, changing the source image or URL is usually more practical than resubmitting the same request.