Currently worker-comfyui only returns generated images (output.images) as Base64 or S3 URLs. Modern ComfyUI workflows (Wan, HunyuanVideo, LTX-Video, CogVideoX, Cosmos, etc.) commonly generate MP4 videos, but these outputs are ignored by the worker.
This PR adds first-class support for video outputs while remaining fully backward compatible.
Motivation
Many recent ComfyUI workflows generate videos instead of images using nodes such as:
VHS Video Combine
SaveVideo
SaveWEBM
FFmpeg Video Output
Other custom video output nodes
The worker currently only scans for image files, making it impossible to use these workflows through the serverless API without custom modifications.
Implementation
Detect video outputs in the ComfyUI output directory.
Support common video extensions:
.mp4
.webm
.mov
.mkv (optional)
Reuse the existing upload pipeline:
Base64 output
S3 upload
Return videos under a new output.videos field.
Preserve existing image behavior with no breaking changes.
Currently worker-comfyui only returns generated images (output.images) as Base64 or S3 URLs. Modern ComfyUI workflows (Wan, HunyuanVideo, LTX-Video, CogVideoX, Cosmos, etc.) commonly generate MP4 videos, but these outputs are ignored by the worker.
This PR adds first-class support for video outputs while remaining fully backward compatible.
Motivation
Many recent ComfyUI workflows generate videos instead of images using nodes such as:
VHS Video Combine
SaveVideo
SaveWEBM
FFmpeg Video Output
Other custom video output nodes
The worker currently only scans for image files, making it impossible to use these workflows through the serverless API without custom modifications.
Implementation
Detect video outputs in the ComfyUI output directory.
Support common video extensions:
.mp4
.webm
.mov
.mkv (optional)
Reuse the existing upload pipeline:
Base64 output
S3 upload
Return videos under a new output.videos field.
Preserve existing image behavior with no breaking changes.