What's Changed
- Improve qwen 2.1 KV cache location logic by @rattus128 in #16429
- Compile Qwen Image 2.1 transformer blocks by @rattus128 in #16430
- Allow model files to contain which attention should be used for each block. by @comfyanonymous in #16419
- JsonExtractString can now handle JSON with prefix and/or suffix. by @comfyanonymous in #16439
- Remove torchaudio dependency. by @comfyanonymous in #16457
- Cuda graphs + memory compiler on ace step 1.5 autoregressive model. by @comfyanonymous in #16461
- Blend H3 VAE tiles against composited neighbours by @kijai in #16436
- Fast disk detection to all model loaders. by @comfyanonymous in #16425
- [Partner Nodes] feat(Tencent): add Hunyuan Image 3.5 text-to-image and edit nodes by @comfyui-wiki in #16462
- Update workflow templates to v0.11.68 by @comfyui-wiki in #16466
- TextGenerate: Add system_prompt input and separate thinking output (CORE-460) by @kijai in #16442
- MiniMax-H3: Support Fun-Controlnet-Union 2.0 (CORE-462) by @kijai in #16471
- Fix MiniMax-H3 VAE rms_rope crash on offloaded qk_norm_scale by @chelsealong in #16485
- Port some optimizations to flux model family. by @comfyanonymous in #16488
- [Partner Nodes] feat(Quiver): add Arrow 2 models with reasoning effort widget by @bigcat88 in #16478
- [Partner Nodes] feat(Anthropic): add Claude Opus 5.5 to the Claude node by @bigcat88 in #16479
- [Partner Nodes] feat(Recraft): add Recraft V4.1 Flash to the V4 text-to-image node by @bigcat88 in #16501
- Update workflow templates to v0.11.69 by @comfyui-wiki in #16503
- Fix ImageUpscaleWithModel crashing on RGBA images by @chelsealong in #16500
- Take the SQLite write lock up front for asset scan and output-registration writes by @synap5e in #16480
- Move file reads out of the remaining asset write transactions by @synap5e in #16486
- feat: ming-image support by @kijai in #16482
- Fix saving/loading ming image model breaking detection. by @comfyanonymous in #16514
- Lower memory usage and .comfy_attention support for lumina family models. by @comfyanonymous in #16515
- [Partner Nodes] feat(OpenAI): add GPT-6 Sol and Luna models by @bigcat88 in #16520
- [Partner Nodes] feat(ByteDance): add Seedream 5.0 Flash by @bigcat88 in #16522
- fix(assets): harden the asset catalogue's failure paths by @synap5e in #16393
- Add missing RDNA2 arches to list. by @comfyanonymous in #16537
- feat(npu): support async weight offload streams by @Big2Wheel in #16057
- Small fix: skip unnecessary work. by @comfyanonymous in #16540
- Fix Qwen VL image preprocessing crash on 4-channel (RGBA) input by @chelsealong in #16548
- [Partner Nodes] feat(ByteDance): add Seedance 2.5 Draft mode by @bigcat88 in #16529
- Update workflow templates to v0.11.70 by @comfyui-wiki in #16557
- Add HDR LogC3 and ACEScct to Convert Color Space node. by @comfyanonymous in #16541
- Fix HDR output issue with negative values in Convert Image Color Space. by @comfyanonymous in #16565
- Update sheetsage2 abc creation from upstream yue code. by @comfyanonymous in #16569
- perf(assets): replace pathlib prefix checks in the asset scan by @synap5e in #16543
- Mention some ComfyUI optimization details in README. by @comfyanonymous in #16576
- Support tiny VAE for Qwen-Image 2.1 by @kijai in #16552
- feat: Support ID-V2V Wan 2.1/VACE based model (CORE-426) by @kijai in #15139
- Fix potential regression with previous PR. by @comfyanonymous in #16596
- [Partner Nodes] chore(Sora): remove deprecated Sora nodes by @bigcat88 in #16609
- Clamp Qwen Image 2.1 fp16 activations in place by @dcbert in #16608
- Add comfy_attention and AttentionTensorContainer to a few models. by @comfyanonymous in #16595
- Speedup Yue2 AR by @rattus128 in #16626
- Searching "kitchen" now returns the attention node. by @comfyanonymous in #16632
- Support Qwen-Image 2.1 union fun controlnet by @kijai in #16519
- Update comfy-kitchen package version to 0.2.36 by @comfyanonymous in #16635
- Optimize Seedvr2 by @kijai in #16530
- Support w6a8 quantization format by @kijai in #16483
- Fixes for last PR. by @comfyanonymous in #16637
- fix(db): wait briefly for the database lock at startup by @synap5e in #16602
- fix(assets): Start without the assets packages; --enable-assets reports what is missing and runs with assets disabled by @synap5e in #16580
- perf(assets): keep the UI responsive during asset scans and rescan output by listing folders by @synap5e in #16546
- Speedup generation on qwen3.5/3.8 with long contexts. by @comfyanonymous in #16638
- Avoid unnecessary write. by @comfyanonymous in #16639
- Update codeowners. by @comfyanonymous in #16655
New Contributors
- @Big2Wheel made their first contribution in #16057
- @dcbert made their first contribution in #16608
Full Changelog: v0.37.0...v0.38.0