What Comfy-Org/ComfyUI shipped
Written by FoxPlug from public releases; not affiliated with ComfyUI. An automatic summary of the public release, pull request and commit data of github.com/Comfy-Org/ComfyUI. ComfyUI did not write it and does not use or endorse FoxPlug. Every line links to the public change it describes.
Get a weekly update like this for your product, free
Week of September 21, 2026
What shipped
- ComfyUI v0.37.4 released. Release
- Added support for Ming-Image model with optimizations for MoE expert handling and quantized weight storage. Pull request #16482
- Added support for ID-V2V Wan 2.1/VACE based model with reference pad image input to I2V node. Pull request #15139
- Added Hunyuan Image 3.5 text-to-image and edit partner API nodes. Pull request #16462
- Reduced memory usage and added
comfy_attentionsupport for lumina family models. Pull request #16515 - Ported memory optimizations and
comfy_attentionsupport to flux model family. Pull request #16488 - Fixed ImageUpscaleWithModel crashing on RGBA images by converting to 3-channel format. Pull request #16500
- Optimized asset scanning performance by replacing slow pathlib prefix checks. Pull request #16543
- Improved asset scan and output registration performance by taking SQLite write lock up front. Pull request #16480
Why it matters
This week shipped model support expansions including Ming-Image and Hunyuan Image 3.5, alongside numerous stability fixes for image handling edge cases and significant performance improvements for asset scanning and database operations. The releases also included memory optimizations across multiple model families and fixes for crashes when processing RGBA images.
Changelog entry
- v0.37.4 released Release
- Added Ming-Image model support with improved MoE expert token grouping and quantized weight storage Pull request #16482
- Added ID-V2V Wan 2.1/VACE model support with reference pad image input for I2V node Pull request #15139
- Added Hunyuan Image 3.5 text-to-image and edit partner API nodes Pull request #16462
- Reduced memory usage and added
comfy_attentionsupport for lumina models Pull request #16515 - Added
comfy_attentionsupport and memory optimizations to flux model family Pull request #16488 - Fixed ImageUpscaleWithModel crashes on RGBA images Pull request #16500
- Fixed HDR output issue with negative values in Convert Image Color Space Pull request #16565
- Added HDR LogC3 and ACEScct color space support Pull request #16541
- Optimized asset scanning performance by replacing pathlib prefix checks Pull request #16543
- Improved asset scan performance by taking SQLite write lock up front Pull request #16480
- TextGenerate node: added
system_promptinput and separate thinking output Pull request #16442
v0.37.4 ships Ming-Image support, Hunyuan Image 3.5 nodes, RGBA image fixes, and asset scanning optimizations. Multiple model families get memory improvements.
ComfyUI v0.37.4 adds Ming-Image support with optimized MoE expert handling, Hunyuan Image 3.5 partner nodes, ID-V2V Wan 2.1 support, and fixes for RGBA image crashes across upscaling and preprocessing. Performance improvements include faster asset scanning through optimized pathlib checks and better SQLite write transaction handling.