Can Comfy Org stop breaking frontend every other update?
Recent ComfyUI update removed key nodes, broke workflows, and cut generation speeds by nearly half.
A recent update to ComfyUI, the popular node-based interface for Stable Diffusion, has sparked significant user frustration due to breaking changes and performance regressions. The update removed the essential 'Flux 2 Conditioning' node and replaced it with a 'Reference Conditioning' node, but failed to maintain backward compatibility. This rendered countless existing workflows unusable. Concurrent UI bugs, such as broken widget rearrangement and copying functions, compounded the instability, leading users to question the platform's update reliability.
In response, many users manually reverted their installations to a previous stable version (backend 0.12.0, frontend 1.39.19). The rollback yielded dramatic results: not only were the UI bugs resolved, but performance saw a substantial boost. One user reported that Flux 2 9B FP8 generation time on an RTX 5060 Ti dropped from 4.20 seconds per iteration on the new version to 2.88 s/it on the old one—a 45% improvement in speed. This performance regression in the new release has been called 'embarrassing' by the community, highlighting a critical failure in the update's quality assurance.
- Update removed 'Flux 2 Conditioning' node without backward compatibility, breaking existing workflows.
- Users reverting to v0.12.0 saw a 45% performance gain (4.20 s/it to 2.88 s/it for Flux 2).
- The release introduced multiple UI bugs, including broken widget rearrangement and copying functions.
Why It Matters
Frequent breaking updates disrupt professional workflows and waste hours of work, undermining trust in essential AI tools.