This release patches 1 CVE for security teams tracking exposure across their dependency inventory.
Topics
ReleasePort's take
Moderate signalComfyUI v0.21.0 removes IPEX (Intel Extension for PyTorch) support entirely.
Why it matters: If your pipelines depend on IPEX acceleration, plan to migrate or adjust configurations before upgrading; no fallback provided.
Summary
AI summaryRemoved IPEX (Intel Extension for PyTorch) support.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Breaking | Medium |
Remove IPEX Intel Extension for Pytorch support Remove IPEX Intel Extension for Pytorch support Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Breaking | Medium |
Remove Moonvalley API nodes from codebase Remove Moonvalley API nodes from codebase Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Breaking | Medium |
Make ColorTransfer node ref_image parameter mandatory Make ColorTransfer node ref_image parameter mandatory Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Auto-regressive video generation support added Auto-regressive video generation support added Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Switch image loading from pillow to pyav library Switch image loading from pillow to pyav library Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Cogvideox video generation model support Cogvideox video generation model support Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Void model pass 1 & 2 implementation added Void model pass 1 & 2 implementation added Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Gemma4 text generation support added Gemma4 text generation support added Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Add causal window fix for context windows blending Add causal window fix for context windows blending Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
dynamicVRAM with --cache-ram 2 memory mode dynamicVRAM with --cache-ram 2 memory mode Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Read audio and video simultaneously in video loader Read audio and video simultaneously in video loader Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Add high quality preview support for Flux2 latents Add high quality preview support for Flux2 latents Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
I2V support for causal forcing video model I2V support for causal forcing video model Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Add Wan-Dancer video generation model support Add Wan-Dancer video generation model support Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
BiRefNet background remove model support BiRefNet background remove model support Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Add generic feature-flag CLI arg and registry Add generic feature-flag CLI arg and registry Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Add description field to blueprint subgraphs Add description field to blueprint subgraphs Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
PrimitiveInt control widget defaults to fixed mode PrimitiveInt control widget defaults to fixed mode Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Enable triton comfy kitchen via CLI argument Enable triton comfy kitchen via CLI argument Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Improve SAM3 large input handling for image processing Improve SAM3 large input handling for image processing Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Performance | Medium |
Block prefetch and LoRA async load implemented for speedup Block prefetch and LoRA async load implemented for speedup Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Performance | Medium |
Reduce video tiny VAE peak VRAM and decode time Reduce video tiny VAE peak VRAM and decode time Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Performance | Medium |
Lower peak memory usage for 8-bit formats with pyav Lower peak memory usage for 8-bit formats with pyav Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Handle uninitialized meta-tensors in models fixes CPU crashes Handle uninitialized meta-tensors in models fixes CPU crashes Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Fix SolidMask and MaskComposite device mismatch with GPU-only Fix SolidMask and MaskComposite device mismatch with GPU-only Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
ImageBlend handles images with different channel counts ImageBlend handles images with different channel counts Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Handle port-in-use error on server startup gracefully Handle port-in-use error on server startup gracefully Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Handle metadata rotation in pyav image loading Handle metadata rotation in pyav image loading Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Return millisecond timestamps from get_file_info API Return millisecond timestamps from get_file_info API Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
LTXVImgToVideoInplace no longer mutates input latents LTXVImgToVideoInplace no longer mutates input latents Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Bugfix | Medium |
Prevent LTXVImgToVideoInplace from mutating input latents and dropping noise_mask Prevent LTXVImgToVideoInplace from mutating input latents and dropping noise_mask Source: granite4.1:30b@2026-05-24-audit Confidence: low |
— |
Full changelog
What's Changed
- Read audio and video at the same time in video loader node. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13591
- fix: Handle un-inited meta-tensors in models (fixes a CPU TE crash) (CORE-67) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13578
- ci: dispatch tag push to Comfy Cloud by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13541
- Change the
save 3d modelnode's filename prefix to3d/ComfyUI(CORE-106) by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/12826 - Fix SolidMask and MaskComposite device mismatch with --gpu-only (CORE-98) by @gschreiber in https://github.com/Comfy-Org/ComfyUI/pull/13296
- Handle metadata rotation in pyav code. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13605
- comfy-aimdo v0.3.0 by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13604
- Make pyav loading code handle tRNS PNG. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13607
- Use pyav to load images instead of pillow (CORE-95) by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13594
- dynamicVRAM + --cache-ram 2 (CORE-117) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13603
- Reduce video tiny VAE peak VRAM and decode time (CORE-127) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13617
- Cogvideox by @Talmaj in https://github.com/Comfy-Org/ComfyUI/pull/13402
- Add high quality preview support for Flux2 latents by @blepping in https://github.com/Comfy-Org/ComfyUI/pull/13496
- Lower peak mem usage for 8 bit formats with pyav. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13626
- Fix SQLAlchemy version format in requirements.txt (CORE-105) by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13547
- [Partner Nodes] allow custom resolutions for GPTImage2 node by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13631
- [Partner Nodes] ByteDance: virtual portrait library for regular images by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13638
- Load other jpeg formats without taking so much memory. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13642
- OneTrainer ERNIE LoRA support by @prookyon in https://github.com/Comfy-Org/ComfyUI/pull/13640
- Update workflow templates to v0.9.65 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13644
- Add alexisrolland and rattus128 as code owners by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13648
- [Partner Nodes] always display the custom width and height in GPTImage2 by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13651
- Reformat models variable into multiline array (CORE-59) by @Talmaj in https://github.com/Comfy-Org/ComfyUI/pull/13513
- chore(api-nodes): remove Moonvalley API nodes by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13659
- Update workflow templates to v0.9.66 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13662
- chore(api-nodes): increase default timeout for partner API node tasks by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13663
- Remove IPEX (Intel Extension for Pytorch) support. by @simonlui in https://github.com/Comfy-Org/ComfyUI/pull/13653
- SDPose: resize fix (CORE-146) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13656
- Adding kijai to codeowners by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13664
- List all the portable downloads in the README section. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13666
- Add script in AMD portable to launch with dynamic vram. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13667
- Implement block prefetch + Lora Async load + and adopt in LTX (Speedup!) (CORE-111) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13618
- Some cleanups to the load image node (CORE-145) by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13677
- Update workflow templates to v0.9.68 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13678
- [Partner Nodes] Topaz Astra 2 model by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13672
- feat: Gemma4 text generation support (CORE-30) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13376
- Update README (CORE-144) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13679
- fix: Make ImageBlend and ImageCompositeMasked nodes handle images with different channel counts (CORE-103) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13615
- Update ComfyUI screenshot in README (CORE-144) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13683
- fix: gracefully handle port-in-use error on server startup by @luke-mino-altherr in https://github.com/Comfy-Org/ComfyUI/pull/13001
- Batch broadcasting in JoinImageWithAlpha node (CORE-143) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13686
- Enable triton comfy kitchen via cli-arg by @silveroxides in https://github.com/Comfy-Org/ComfyUI/pull/12730
- Refactor LoadImageMask to use LoadImage code (CORE-142) by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13687
- PrimitiveInt: default control_after_generate widget to fixed (CORE-141) by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13690
- fix: Make ColorTransfer node ref_image mandatory (CORE-140) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13691
- Fix a1111 typo in extra_model_paths.yaml by @soof-golan in https://github.com/Comfy-Org/ComfyUI/pull/2720
- fix: Proper memory estimation for frame interpolation (CORE-150) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13698
- prefetch: guard against no offload (CORE-149) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13703
- make control-net load order deterministic (CORE-139) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13701
- ops: handle multi-compute of the same weight (CORE-153) by @rattus128 in https://github.com/Comfy-Org/ComfyUI/pull/13705
- chore: Update display names and categories (CORE-151) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13693
- fix(spec): mark DeviceStats.index and NodeInfo.essentials_category as nullable by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13706
- spec: add workflow_id / workflow_version_id to PromptRequest with x-runtime tag by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13709
- feat: add generic --feature-flag CLI arg and --list-feature-flags registry by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13685
- Add deploy environment header (Comfy-Env) to partner node API calls by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13425
- Document core stable release frequency is now ~2 weeks. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13710
- feat: Auto-regressive video generation (CORE-25) by @Talmaj in https://github.com/Comfy-Org/ComfyUI/pull/13082
- fix: SplitImageToTileList and ImageMergeTileList to use tile_height for vertical stride minimum by @alvinttang in https://github.com/Comfy-Org/ComfyUI/pull/12882
- [Partner Nodes] add Luma UNI-1 model by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13614
- [Partner Nodes] add Gpt 5.5 and 5.5-pro LLM models by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13673
- Update workflow templates to v0.9.69 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13714
- chore: Update display names and categories for text nodes (CORE-155) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13712
- Fix Content-Disposition header missing attachment prefix by @themachinehf in https://github.com/Comfy-Org/ComfyUI/pull/13093
- fix: make VAEDecodeAudio usable for LTX-2.x generated audio latents (CORE-157) by @drozbay in https://github.com/Comfy-Org/ComfyUI/pull/13716
- Add Spectral lint CI gate for openapi.yaml by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13410
- fix(api-io): serialize MultiCombo multi_select as object config by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13484
- feat: add cloud-specific fields to OSS openapi.yaml as nullable by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13623
- Add temporal_downscale_ratio to LatentFormat (CORE-147) by @drozbay in https://github.com/Comfy-Org/ComfyUI/pull/13702
- feat: Context windows - add causal_window_fix to improve blending of context windows (CORE-100) by @drozbay in https://github.com/Comfy-Org/ComfyUI/pull/13563
- First step of supporting save filenames without trailing _ by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13722
- Update language options in nodes_ace.py by @iChristGit in https://github.com/Comfy-Org/ComfyUI/pull/12578
- fix: return millisecond timestamps from get_file_info() by @luke-mino-altherr in https://github.com/Comfy-Org/ComfyUI/pull/12996
- Void model - pass 1 & 2 (CORE-38) by @Talmaj in https://github.com/Comfy-Org/ComfyUI/pull/13403
- Bump comfyui-frontend-package to 1.43.17 by @comfy-pr-bot in https://github.com/Comfy-Org/ComfyUI/pull/13723
- Revert "Fix Content-Disposition header missing 'attachment;' prefix (#13093)" by @guill in https://github.com/Comfy-Org/ComfyUI/pull/13733
- Throttle dynamic VRAM prepare logging by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13704
- Add .comfy_environment to portable. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13746
- [Partner Nodes] add grok-imagine-image-quality model by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13725
- Update workflow templates to v0.9.72 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13732
- [Partner Nodes] fix of price badge for Kling V3 model in the Motion Control by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13790
- [Partner Nodes] new NanoBanana2 node with DynamicCombo/Autogrow by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13753
- Improve SAM3 large input handling (CORE-174) by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13767
- I2V for causal forcing model (CORE-26) by @Talmaj in https://github.com/Comfy-Org/ComfyUI/pull/13719
- Update warning message for ComfyUI frontend installation. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13796
- fix: make NodeReplaceManager.register() idempotent (CORE-171) by @Kosinkadink in https://github.com/Comfy-Org/ComfyUI/pull/13596
- Update nodes categories and display names (CORE-89) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13786
- [Partner Nodes] use "adaptive" aspect ratio for SD2 nodes by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13800
- Fix typos by @omahs in https://github.com/Comfy-Org/ComfyUI/pull/10986
- Add support for BiRefNet background remove model (CORE-46) by @yousef-rafat in https://github.com/Comfy-Org/ComfyUI/pull/12747
- fix: Make LTXVAddGuide center-crop guide images to match other LTXV nodes (CORE-182) by @drozbay in https://github.com/Comfy-Org/ComfyUI/pull/13794
- fix: Stop LTXVImgToVideoInplace from mutating input latents and dropping noise_mask (CORE-183) by @drozbay in https://github.com/Comfy-Org/ComfyUI/pull/13793
- fix: Fix bug when mask not on same device (CORE-181) by @alexisrolland in https://github.com/Comfy-Org/ComfyUI/pull/13801
- Add cloud-runtime FE-facing operations to spec by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13734
- [Partner Nodes] additionally use Baidu to detect the accessibility of internet by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13803
- I don't think there was any because nobody complained. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13807
- Add cloud-runtime experiment node-schema endpoints to spec by @MillerMedia in https://github.com/Comfy-Org/ComfyUI/pull/13806
- Add description field to blueprint subgraphs by @lin-bot23 in https://github.com/Comfy-Org/ComfyUI/pull/13797
- [Partner Nodes] Tripo3D 3.1 model by @bigcat88 in https://github.com/Comfy-Org/ComfyUI/pull/13788
- Bump comfyui-frontend-package to 1.43.18 by @comfy-pr-bot in https://github.com/Comfy-Org/ComfyUI/pull/13809
- Support Wan-Dancer by @kijai in https://github.com/Comfy-Org/ComfyUI/pull/13813
- Use LatentCutToBatch instead. by @comfyanonymous in https://github.com/Comfy-Org/ComfyUI/pull/13815
- Add Boolean support to Math Expression Node by @LaVie024 in https://github.com/Comfy-Org/ComfyUI/pull/13224
- Update workflow templates to v0.9.73 by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13822
- Remove advanced flag from layers input in EmptyQwenImageLayeredLatent node by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13823
- Add New Blueprints by @comfyui-wiki in https://github.com/Comfy-Org/ComfyUI/pull/13570
- chore: remove extra word in comment by @box4wangjing in https://github.com/Comfy-Org/ComfyUI/pull/13826
New Contributors
- @gschreiber made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/13296
- @prookyon made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/13640
- @soof-golan made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/2720
- @alvinttang made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/12882
- @themachinehf made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/13093
- @iChristGit made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/12578
- @omahs made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/10986
- @lin-bot23 made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/13797
- @box4wangjing made their first contribution in https://github.com/Comfy-Org/ComfyUI/pull/13826
Full Changelog: https://github.com/Comfy-Org/ComfyUI/compare/v0.20.1...v0.21.0
Breaking Changes
- Removed IPEX (Intel Extension for PyTorch) support.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About ComfyUI
The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.
Related context
Related tools
Earlier breaking changes
- v0.21.1 Revert breaking changes from previous release
Beta — feedback welcome: [email protected]