What's Changed

• [API Nodes] GPT-Image-1.5 by @bigcat88 in #11368
• fix regression in V3 nodes processing by @bigcat88 in #11375
• SA-Solver: return the result of the last predictor step correctly by @chaObserv in #11380
• Better handle torch being imported by prestartup nodes. by @comfyanonymous in #11383

Full Changelog: v0.5.0...v0.5.1

What's Changed

• Bump frontend to 1.34.8 by @benceruleanlu in #11220
• fix: correctly process the Node V1 dict results on V3 node by @bigcat88 in #11237
• Tweak Z Image memory estimation. by @comfyanonymous in #11254
• Fix: filter hidden files from /internal/files endpoint by @Myestery in #11191
• Lower VAE loading requirements：Create a new branch for GPU memory calculations in qwen-image vae by @jiangchengchengark in #11199
• feat(api-nodes): enable Kling Omni Image node by @bigcat88 in #11229
• Adjust memory usage factor for sd3. by @comfyanonymous in #11257
• Fix regular chroma radiance by @comfyanonymous in #11276
• This only works on radiance. by @comfyanonymous in #11277
• Better chroma radiance vram estimation. by @comfyanonymous in #11278
• Make portable updater work with repos in unmerged state. by @comfyanonymous in #11281
• WanMove support by @kijai in #11247
• Respect the dtype the op was initialized in for non quant mixed op. by @comfyanonymous in #11282
• feat(api-nodes): Kling TextToVideoWithAudio and ImageToVideoWithAudio nodes by @bigcat88 in #11267
• Fix bias dtype issue in mixed ops. by @comfyanonymous in #11293
• Basic implementation of z image fun control union 2.0 by @comfyanonymous in #11304
• Fix pytorch warnings. by @comfyanonymous in #11314
• seeds_2: add phi_2 variant and sampler node by @chaObserv in #11309
• Update warning for old pytorch version. by @comfyanonymous in #11319
• bump manager to the 4.0.3b5 by @ltdrdata in #11324
• Update workflow templates to v0.7.59 by @comfyui-wiki in #11337
• Disable guards on transformer_options when torch.compile by @comfyanonymous in #11317
• [API Nodes] drop very old Kling models by @bigcat88 in #11307
• drop Pika API nodes by @bigcat88 in #11306
• feat(preview): add per-queue live preview method override by @ltdrdata in #11261
• comfy_api: remove usage of "Type","List" and "Dict" types by @bigcat88 in #11238
• Add context windows callback for custom cond handling by @drozbay in #11208
• Support the new qwen edit 2511 reference method. by @comfyanonymous in #11340
• [add] tripo3.0 by @seed93 in #10663
• Add code to detect if a z image fun controlnet is broken or not. by @comfyanonymous in #11341
• [BlockInfo] Lumina by @Haoming02 in #11227
• [BlockInfo] Wan by @Haoming02 in #10845
• Only enable fp16 on ZImage on newer pytorch. by @comfyanonymous in #11344
• Patch frontend to 1.34.9 (from 1.34.8) by @christian-byrne in #11342
• Inpainting for z image fun control. Use the ZImageFunControlnet node. by @comfyanonymous in #11346
• Add a way to set the default ref method in the qwen image code. by @comfyanonymous in #11349
• Update workflows for new release process by @benceruleanlu in #11064
• [API Nodes] add Wan2.6 model to video nodes by @bigcat88 in #11357
• Check state dict key to auto enable the index_timestep_zero ref method. by @comfyanonymous in #11362
• Add exp_heun_2_x0 sampler series by @chaObserv in #11360

New Contributors

• @jiangchengchengark made their first contribution in #11199
• @drozbay made their first contribution in #11208
• @seed93 made their first contribution in #10663

Full Changelog: v0.4.0...v0.5.0

What's Changed

• Implement the Ovis image model. by @comfyanonymous in #11030
• Bump frontend to 1.33 by @christian-byrne in #11028
• feat: Support ComfyUI-Manager for pip version by @ltdrdata in #7555
• Add @guill as a code owner by @yoland68 in #11031
• Fix CODEOWNERS formatting to have all on the same line by @Kosinkadink in #11053
• add check for the format arg type in VideoFromComponents.save_to function by @bigcat88 in #11046
• Hack to make zimage work in fp16. by @comfyanonymous in #11057
• attention: use flag based OOM fallback by @rattus128 in #11038
• Support Z Image alipai controlnets. by @comfyanonymous in #11062
• Added PATCH method to CORS headers by @jheising in #11066
• Implement temporal rolling VAE (Major VRAM reductions in Hunyuan and Kandinsky) by @rattus128 in #10995
• Add MatchType, DynamicCombo, and Autogrow support to V3 Schema by @Kosinkadink in #10832
• Fix issue with portable updater. by @comfyanonymous in #11070
• Prs/lora reservations (reduce massive Lora reservations especially on Flux2) by @rattus128 in #11069
• use empty list defaults for Schema.inputs/outputs/hidden to avoid None issues by @bigcat88 in #11083
• [API nodes] add support for "@image" reference format in Kling Omni API nodes by @bigcat88 in #11082
• [V3] convert nodes_load_3d.py to V3 schema by @bigcat88 in #10990
• [V3] convert nodes_audio.py to V3 schema by @bigcat88 in #10798
• Qwen Image Lora training fix from #11090 by @comfyanonymous in #11094
• Fix case where text encoders where running on the CPU instead of GPU. by @comfyanonymous in #11095
• mp: use look-ahead actuals for stream offload VRAM calculation (fixes unwanted TE full offload) by @rattus128 in #11096
• sd: revise hy VAE VRAM by @rattus128 in #11105
• sd: bump HY1.5 VAE estimate by @rattus128 in #11107
• [API Nodes]: fixes and refactor by @bigcat88 in #11104
• Forgot to put this in README. by @comfyanonymous in #11112
• Fix transformer_options getting cleared in Lumina model after z-image PR by @Kosinkadink in #11118
• Make old scaled fp8 format use the new mixed quant ops system. by @comfyanonymous in #11000
• Fix regression when text encoder loaded directly on GPU. by @comfyanonymous in #11129
• Context windows fixes and features by @kijai in #10975
• Fix some custom nodes. by @comfyanonymous in #11134
• docs: add ComfyUI-Manager documentation and update to v4.0.3b4 by @ltdrdata in #11133
• Kandinsky5 model support by @kijai in #10988
• Fix regression. by @comfyanonymous in #11137
• [V3] convert nodes_freelunch.py to the V3 schema by @bigcat88 in #10904
• [V3] convert nodes_mask.py to V3 schema by @bigcat88 in #10669
• Set OCL_SET_SVM_SIZE on AMD. by @comfyanonymous in #11139
• marked all Pika API nodes a deprecated by @bigcat88 in #11146
• Bugfix: Wrong EmptyAudio node input types by @kijai in #11149
• Speed up lora compute and lower memory usage by doing it in fp16. by @comfyanonymous in #11161
• Fix on-load VRAM OOM by @rattus128 in #11144
• Fix qwen scaled fp8 not working with kandinsky. Make basic t2i wf work. by @comfyanonymous in #11162
• Properly load the newbie diffusion model. by @comfyanonymous in #11172
• chore(comfy_api): replace absolute imports with relative by @bigcat88 in #11145
• Update workflow templates to v0.7.51 by @comfyui-wiki in #11150
• chore: replace imports of deprecated V1 classes by @bigcat88 in #11127
• [API Nodes] Add System Instruction support & Visual Enforcement for Banana Nodes by @bigcat88 in #11177
• [API Nodes] add support for seedance-1-0-pro-fast model by @bigcat88 in #10947
• Support "transformer." LoRA prefix for Z-Image by @dxqb in #11135
• retune lowVramPatch VRAM accounting by @rattus128 in #11173
• Update workflow templates to v0.7.54 by @comfyui-wiki in #11192
• Fix regression. by @comfyanonymous in #11194
• Fix potential issue. by @comfyanonymous in #11201
• dequantization offload accounting (fixes Flux2 OOMs - incl TEs) by @rattus128 in #11171
• Patch frontend to 1.33.13 (from 1.33.10) by @christian-byrne in #11200
• add chroma-radiance-x0 mode by @lodestone-rock in #11197
• ops: delete dead code by @rattus128 in #11204
• Fix for HunyuanVideo1.5 meanflow distil by @kijai in #11212
• Fix nan issue when quantizing fp16 tensor. by @comfyanonymous in #11213

New Contributors

• @jheising made their first contribution in #11066
• @dxqb made their first contribution in #11135
• @lodestone-rock made their first contribution in #11197

Full Changelog: v0.3.76...v0.4.0

What's Changed

• Add cheap latent preview for flux 2. by @comfyanonymous in #10907
• [API Nodes] add Veo3 First-Last-Frame node by @bigcat88 in #10878
• [API Nodes] improve UX for batch uploads in upload_images_to_comfyapi by @bigcat88 in #10913
• [API Nodes] fix(gemini): use first 10 images as fileData (URLs) and remaining images as inline base64 by @bigcat88 in #10918
• Bump frontend to 1.32.9 by @christian-byrne in #10867
• Merge 3d animation node by @jtydhr88 in #10025
• Fix the CSP offline feature on latest frontend. by @comfyanonymous in #10923
• Add Z Image to readme. by @comfyanonymous in #10924
• chore(api-nodes): remove chat widgets from OpenAI/Gemini nodes by @bigcat88 in #10861
• [V3] convert nodes_custom_sampler.py to V3 schema by @bigcat88 in #10206
• Dataset Processing Nodes and Improved LoRA Trainer Nodes with multi resolution supports. by @KohakuBlueleaf in #10708
• Make lora training work on Z Image and remove some redundant nodes. by @comfyanonymous in #10927
• [BlockInfo] Flux by @Haoming02 in #10841
• Account for the VRAM cost of weight offloading by @rattus128 in #10733
• quant ops: Dequantize weight in-place (reduce flux2 VRAM usage) by @rattus128 in #10935
• Update template to 0.7.23 by @comfyui-wiki in #10949
• Enable async offloading by default on Nvidia. by @comfyanonymous in #10953
• feat(Kling-API-Nodes): add v2-5-turbo model to FirstLastFrame node by @bigcat88 in #10938
• fix(user_manager): fix typo in move_userdata dest validation by @ltdrdata in #10967
• Disable offload stream when torch compile. by @comfyanonymous in #10961
• fix QuantizedTensor.is_contiguous (#10956) by @urlesistiana in #10959
• mm: wrap the raw stream in context manager by @rattus128 in #10958
• Update driver link in AMD portable README by @comfyanonymous in #10974
• Support video tiny VAEs by @kijai in #10884
• Support some z image lora formats. by @comfyanonymous in #10978
• feat(security): add System User protection with __ prefix by @ltdrdata in #10966
• Add missing z image lora layers. by @comfyanonymous in #10980
• Make the ScaleRope node work on Z Image and Lumina. by @comfyanonymous in #10994
• update template to 0.7.25 by @comfyui-wiki in #10996
• Next AMD portable will have pytorch with ROCm 7.1.1 by @comfyanonymous in #11002
• Bumps frontend to 1.32.10 (from 1.32.9) by @christian-byrne in #11018
• Update qwen tokenizer to add qwen 3 tokens. by @comfyanonymous in #11029
• [API Nodes] add Kling O1 model support by @bigcat88 in #11025

New Contributors

• @urlesistiana made their first contribution in #10959

Full Changelog: v0.3.75...v0.3.76

What's Changed

• Z Image model. by @comfyanonymous in #10892
• Adjustments to Z Image. by @comfyanonymous in #10893
• Fix loras not working on mixed fp8. by @comfyanonymous in #10899
• Fix Flux2 reference image mem estimation. by @comfyanonymous in #10905

Full Changelog: v0.3.73...v0.3.75

What's Changed

• Fix crash. by @comfyanonymous in #10885
• Update workflow templates to v0.7.20 by @comfyui-wiki in #10883
• Lower vram usage for flux 2 text encoder. by @comfyanonymous in #10887

Full Changelog: v0.3.72...v0.3.73

What's Changed

• Bump frontend to 1.30.6 by @christian-byrne in #10793
• --disable-api-nodes now sets CSP header to force frontend offline. by @comfyanonymous in #10829
• update workflow templates (to add hunyuan video and nano banana pro variants) by @christian-byrne in #10834
• Add display names to Hunyuan latent video nodes. by @comfyanonymous in #10837
• Add better error message for common error. by @comfyanonymous in #10846
• [fix] Fixes non-async public API access by @guill in #10857
• fix(api-nodes): edge cases in responses for Gemini models by @bigcat88 in #10860
• add get_frame_count and get_frame_rate methods to VideoInput class by @bigcat88 in #10851
• [BlockInfo] Chroma by @Haoming02 in #10843
• [BlockInfo] Qwen-Image by @Haoming02 in #10842
• [BlockInfo] HunyuanVideo by @Haoming02 in #10844
• Bump transformers version in requirements.txt by @comfyanonymous in #10869
• Cleanup and fix issues with text encoder quants. by @comfyanonymous in #10872
• Allow pinning quantized tensors. by @comfyanonymous in #10873
• Don't try fp8 matrix mult in quantized ops if not supported by hardware. by @comfyanonymous in #10874
• I found a case where this is needed by @comfyanonymous in #10875
• Flux 2 by @comfyanonymous in #10879
• [API Nodes] add Flux.2 Pro node by @bigcat88 in #10880
• Add Flux 2 support to README. by @comfyanonymous in #10882

Full Changelog: v0.3.71...v0.3.72

What's Changed

• Add a way to disable the final norm in the llama based TE models. by @comfyanonymous in #10794
• change display name of PreviewAny node to "Preview as Text" by @bigcat88 in #10796
• [V3] convert hunyuan3d.py to V3 schema by @bigcat88 in #10664
• Fix workflow name. by @comfyanonymous in #10806
• [API Nodes] add Topaz API nodes by @bigcat88 in #10755
• Disable workaround on newer cudnn. by @comfyanonymous in #10807
• Update server templates handler to use new multi-package distribution (comfyui-workflow-templates versions >=0.3) by @christian-byrne in #10791
• Fix ImageBatch with different channel count. by @comfyanonymous in #10815
• Make Batch Images node add alpha channel when one of the inputs has it by @Kosinkadink in #10816
• feat(api-nodes): add Nano Banana Pro by @bigcat88 in #10814
• fix(KlingLipSyncAudioToVideoNode): convert audio to mp3 format by @bigcat88 in #10811
• bump comfyui-workflow-templates for nano banana 2 by @christian-byrne in #10818
• HunyuanVideo 1.5 by @comfyanonymous in #10819
• Fix wrong path. by @comfyanonymous in #10821

Full Changelog: v0.3.70...v0.3.71

What's Changed

• Add release workflow for NVIDIA cu126 by @comfyanonymous in #10777
• Update README with new portable download link by @comfyanonymous in #10778
• Fix the portable download link for CUDA 12.6 by @comfyui-wiki in #10780
• Native block swap custom nodes considered harmful. by @comfyanonymous in #10783
• [API nodes]: adjusted PR template; set min python version for pylint to 3.10 by @bigcat88 in #10787
• EasyCache: Fix for mismatch in input/output channels with some models by @kijai in #10788
• Fix hunyuan 3d 2.0 by @comfyanonymous in #10792
• feat(api-nodes): add new Gemini models by @bigcat88 in #10789

Full Changelog: v0.3.69...v0.3.70

What's Changed

• Use single apply_rope function across models by @contentis in #10547
• Lower ltxv mem usage to what it was before previous pr. by @comfyanonymous in #10643
• feat(API-nodes): use new client in Rodin3D nodes; remove old api client by @bigcat88 in #10645
• Fix qwen controlnet regression. by @comfyanonymous in #10657
• Enable pinned memory by default on Nvidia. by @comfyanonymous in #10656
• Pinned mem also seems to work on AMD. by @comfyanonymous in #10658
• Clarify release cycle. by @comfyanonymous in #10667
• Tell users they need to upload their logs in bug reports. by @comfyanonymous in #10671
• mm: guard against double pin and unpin explicitly by @rattus128 in #10672
• Only unpin tensor if it was pinned by ComfyUI by @comfyanonymous in #10677
• Make ScaleROPE node work on Flux. by @comfyanonymous in #10686
• Add logging for model unloading. by @comfyanonymous in #10692
• Unload weights if vram usage goes up between runs. by @comfyanonymous in #10690
• ops: Put weight cast on the offload stream - Fixes --async-offload black screen by @rattus128 in #10697
• Update CI workflow to remove dead macOS runner. by @comfyanonymous in #10704
• Don't pin tensor if not a torch.nn.parameter.Parameter by @comfyanonymous in #10718
• Update README.md for Intel Arc GPU installation, remove IPEX by @qiacheng in #10729
• always unload re-used but modified models - Fixed bad outputs in some Upscaler / Lora flows by @rattus128 in #10724
• qwen: reduce VRAM usage by @rattus128 in #10725
• Update Python 3.14 compatibility notes in README by @comfyanonymous in #10730
• Quantized Ops fixes by @contentis in #10715
• add PR template for API-Nodes by @bigcat88 in #10736
• feat: add create_time dict to prompt field in /history and /queue by @ric-yu in #10741
• flux: reduce VRAM usage by @rattus128 in #10737
• Better instructions for the portable. by @comfyanonymous in #10743
• Use same code for chroma and flux blocks so that optimizations are shared. by @comfyanonymous in #10746
• Fix custom nodes import error. by @comfyanonymous in #10747
• Add left padding support to tokenizers. by @comfyanonymous in #10753
• [API Nodes] mark OpenAIDalle2 and OpenAIDalle3 nodes as deprecated by @bigcat88 in #10757
• Revert "mark OpenAIDalle2 and OpenAIDalle3 nodes as deprecated (#10757)" by @bigcat88 in #10759
• Change ROCm nightly install command to 7.1 by @comfyanonymous in #10764

New Contributors

• @ric-yu made their first contribution in #10741

Full Changelog: v0.3.68...v0.3.69