The efficiency issue of batch processing.
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 158
Description
Workflow Description: Batch image generation using ControlNet's OpenPose (approximately 100 poses). After KSampler, images go through resizing, facial repair, hand repair, and are finally saved.
Execution Process: Execution failed due to insufficient VRAM. Reduced the batch size (poses) to 50, which allowed execution but at a much slower speed, taking twice the time compared to the web UI. Upon observing the processing flow, I noticed that ComfyUI processes 50 poses together, resizing, repairing, and saving them in parallel, which leads to VRAM depletion and reduced efficiency. In contrast, the web UI processes images sequentially, one after the other.
Expectation: Is there a node or a workflow that can achieve sequential operations similar to the web UI?
Contributor guide
Assessment
This issue has not been assessed yet.