v0.20.1

What's Changed

• feat: SUPIR model support (CORE-17) by @kijai in #13250
• Some optimizations to make Ernie inference a bit faster. by @comfyanonymous in #13472
• fix: append directory type annotation to internal files endpoint (CORE-71) by @Abdulrehman-PIAIC80387 in #13305
• Add link to Intel portable to Readme. by @comfyanonymous in #13477
• fix: pin SQLAlchemy>=2.0 in requirements.txt (CORE-63) by @octo-patch in #13316
• Refactor io to IO in nodes_ace.py by @comfyanonymous in #13485
• Bump comfyui-frontend-package to 1.42.12 by @comfy-pr-bot in #13489
• Make the ltx audio vae more native (CORE-76) by @comfyanonymous in #13486
• [Partner Nodes] add optional auto-downscaling of videos for ByteDance 2 nodes by @bigcat88 in #13465
• Support standalone LTXV audio VAEs by @kijai in #13499
• [Partner Nodes] added 4K resolution for Veo models; added Veo 3 Lite model by @bigcat88 in #13330
• Bump comfyui-frontend-package to 1.42.14 by @comfy-pr-bot in #13493
• Add gpt-image-2 as version option by @AustinMroz in #13501
• Allow logging in comfy app files. by @comfyanonymous in #13505
• Update workflow templates to v0.9.59 by @comfyui-wiki in #13507
• fix(veo): reject 4K resolution for veo-3.0 models by @MillerMedia in #13504
• feat: RIFE and FILM frame interpolation model support (CORE-29) by @kijai in #13258
• fix: use Parameter assignment for Stable_Zero123 cc_projection weights by @octo-patch in #13518
• Derive InterruptProcessingException from BaseException by @blepping in #13523
• bump manager version to 4.2.1 by @ltdrdata in #13516
• ModelPatcherDynamic: force cast stray weights on comfy layers by @rattus128 in #13487
• Update logging level for invalid version format by @comfyanonymous in #13526
• [Partner Nodes] add SD2 real human support by @bigcat88 in #13509
• feat: SAM (segment anything) 3.1 support (CORE-34) by @kijai in #13408
• [Partner Nodes] GPTImage: fix price badges, add new resolutions by @bigcat88 in #13519
• Update workflow templates to v0.9.61 by @comfyui-wiki in #13533
• Update embedded docs to v0.4.4 by @comfyui-wiki in #13535
• [Partner Nodes] add 4K resolution to Kling nodes by @bigcat88 in #13536
• Fix LTXV Reference Audio node by @kijai in #13531
• comfy-aimdo 0.2.14: Hotfix async allocator estimations by @rattus128 in #13534
• Disable sageattention for SAM3 by @kijai in #13529
• execution: Add anti-cycle validation by @rattus128 in #13169
• Update workflow templates to v0.9.62 by @comfyui-wiki in #13539
• Print more tensor values in the preview any node. by @comfyanonymous in #13544
• fix: use textureSize instead of u_resolution for texel size in blur/sharpen shaders by @jtydhr88 in #13347
• range type by @jtydhr88 in #13322
• Add OpenAPI 3.1 specification for ComfyUI API by @MillerMedia in #13397
• chore(api-nodes): add upcoming-deprecation notice to Sora nodes by @bigcat88 in #13549
• Properly load higher bit depth videos. by @comfyanonymous in #13542
• Bump comfyui-frontend-package to 1.42.15 by @comfy-pr-bot in #13556
• Support loading the alpha channel of videos. by @comfyanonymous in #13564
• Add new ComfyUI blueprints and fix subgraph naming by @comfyui-wiki in #13371
• Make EmptySD3LatentImage node use intermediate dtype. by @comfyanonymous in #13577
• [Partner Nodes] HappyHorse model by @bigcat88 in #13582
• Update workflow templates to v0.9.63 by @comfyui-wiki in #13586

New Contributors

• @Abdulrehman-PIAIC80387 made their first contribution in #13305
• @octo-patch made their first contribution in #13316

Full Changelog: v0.19.3...v0.20.1**Full Changelog**: v0.20.0...v0.20.1

v0.19.3

What's Changed

• nodes_textgen: Implement use_default_template for LTX by @rattus128 in #13451
• fix(api-nodes): corrected StabilityAI price badges by @bigcat88 in #13454
• [Partner Nodes] add new "arrow-1.1" and "arrow-1.1-max" SVG models by @bigcat88 in #13447
• Update workflow templates to v0.9.57 by @comfyui-wiki in #13455
• [Partner Nodes] make "obj" output optional in Hunyuan3D Text and Image to 3D by @bigcat88 in #13449

Full Changelog: v0.19.2...v0.19.3

v0.19.2

What's Changed

• Fix OOM regression in _apply() for quantized models during inference by @jkyamog in #13372
• Add a way to disable default template in text gen node. by @comfyanonymous in #13424
• Use ErnieTEModel_ not ErnieTEModel. by @bedovyy in #13431
• [Partner Nodes] add 1080p resolution for SeeDance 2.0 model by @bigcat88 in #13437
• Add JsonExtractString node. by @comfyanonymous in #13435

New Contributors

• @jkyamog made their first contribution in #13372
• @bedovyy made their first contribution in #13431

Full Changelog: v0.19.1...v0.19.2

v0.19.1

What's Changed

• Fix issue with ernie image. by @comfyanonymous in #13393
• Make text generation work with ministral model. by @comfyanonymous in #13395
• [Partner Nodes] add Sonilo Audio nodes by @bigcat88 in #13391
• Update workflow templates to v0.9.50 by @comfyui-wiki in #13399
• Bump comfyui-frontend-package to 1.42.11 by @comfy-pr-bot in #13398
• Add string output to preview text node. by @comfyanonymous in #13406
• Update workflow templates to v0.9.54 by @comfyui-wiki in #13412
• Fix ernie on devices that don't support fp64. by @comfyanonymous in #13414

Full Changelog: v0.19.0...v0.19.1

v0.19.0

What's Changed

• LongCat-Image edit by @Talmaj in #13003
• Add warning for users who disable dynamic vram. by @comfyanonymous in #13113
• bump manager version to 4.1b8 by @ltdrdata in #13108
• feat: LTX2: Support reference audio (ID-LoRA) (CORE-16) by @kijai in #13111
• Update frontend version to 1.42.8 by @comfyanonymous in #13126
• fix: set default values of Color Adjustment node to zero by @kaili-yang in #13084
• feat(api-nodes): update xAI Grok nodes by @bigcat88 in #13140
• Update templates package version. by @comfyanonymous in #13141
• CURVE node by @jtydhr88 in #12757
• Add Number Convert node by @dante01yoon in #13041
• [Trainer/Experimental] Support fp8 backward by @KohakuBlueleaf in #13121
• feat(assets): register output files as assets after prompt execution by @luke-mino-altherr in #12812
• Fix Train LoRA crash with training_dtype "none" and bfloat16 LoRA weights by @Krishnachaitanyakc in #13145
• [API Nodes] made Reve node price badges more precise by @bigcat88 in #13154
• bump manager version to 4.1 by @ltdrdata in #13156
• fix(number-convert): preserve int precision for large numbers by @dante01yoon in #13147
• CORE-15 feat: Support Qwen3.5 text generation models by @kijai in #12771
• fix(assets): recognize temp directory in asset category resolution by @luke-mino-altherr in #13159
• Make Qwen 8B work with TextGenerate node. by @comfyanonymous in #13160
• fix: regression in text generate with LTXAV model by @kijai in #13170
• [API Nodes] add new Topaz model by @bigcat88 in #13175
• Update workflow templates to v0.9.38 by @comfyui-wiki in #13176
• feat: add curve inputs and raise uniform limit for GLSL shader node by @jtydhr88 in #13158
• Fix blur and sharpen nodes not working with fp16 intermediates. by @comfyanonymous in #13181
• fix: register image/svg+xml MIME type for .svg files by @viva-jinyi in #13186
• Update workflow templates to v0.9.39 by @comfyui-wiki in #13196
• fix(api-nodes-nanobana): raise error when not output image is present by @bigcat88 in #13167
• Allow flux conditioning without a pooled output. by @comfyanonymous in #13198
• Add has_intermediate_output flag for nodes with interactive UI by @jtydhr88 in #13048
• Integrate RAM cache with model RAM management by @rattus128 in #13173
• fix: avoid nested sampler function calls in Color Curves shader by @jtydhr88 in #13209
• Fix #13214 by @comfyanonymous in #13216
• feat: Support RT-DETRv4 detection model (CORE-13) by @kijai in #12748
• [Partner Nodes] Tencent3D update by @bigcat88 in #13207
• Fix/tweak pinned memory accounting by @rattus128 in #13221
• Update README.md with latest AMD Linux pytorch. by @comfyanonymous in #13228
• Rename utils/string nodes with Text prefix and add search aliases by @christian-byrne in #13227
• Update README with note on model support by @comfyanonymous in #13235
• Fix some fp8 scaled checkpoints no longer working. by @comfyanonymous in #13239
• Update workflow templates to v0.9.41 by @comfyui-wiki in #13242
• Remove dead code. by @comfyanonymous in #13251
• Fix some issue with insecure browsers. by @comfyanonymous in #13261
• [API Nodes] add new nodes for WAN2.7 by @bigcat88 in #13264
• Update template to 0.9.43 by @comfyui-wiki in #13265
• Add portable release for intel XPU. by @comfyanonymous in #13272
• Nightly Nvidia pytorch is now cu132 by @comfyanonymous in #13288
• Update workflow templates to v0.9.44 by @comfyui-wiki in #13290
• Update README.md with new frontend release cycle. by @comfyanonymous in #13301
• image histogram node by @jtydhr88 in #13153
• Ace step empty latent nodes follow intermediate dtype. by @comfyanonymous in #13313
• format blueprint by @jtydhr88 in #13315
• feat: Support Ace Step 1.5 XL model (CORE-39) by @comfyanonymous in #13317
• Add support for small flux.2 decoder by @huemin-art in #13314
• Basic intel standalone package .bat by @comfyanonymous in #13333
• Bump comfyui-frontend-package version to 1.42.10 by @comfyanonymous in #13346
• should use 0 as defalut for brightness by @jtydhr88 in #13345
• [Partner Nodes] fix GrokVideoReferenceNode price badge by @bigcat88 in #13354
• Update workflow templates to v0.9.45 by @comfyui-wiki in #13353
• fix: SDPose - resize input always by @kijai in #13349
• Make ImageUpscaleWithModel node work with intermediate device and dtype. by @comfyanonymous in #13357
• Add a supports_fp64 function. by @comfyanonymous in #13368
• Implement Ernie Image model. by @comfyanonymous in #13369
• Update quant doc so it's not completely wrong. by @comfyanonymous in #13381
• Fix typo. by @comfyanonymous in #13382
• [Partner Nodes] add SeeDance 2.0 nodes by @bigcat88 in #13364
• Update workflow templates to v0.9.47 by @comfyui-wiki in #13385

New Contributors

• @kaili-yang made their first contribution in #13084
• @Krishnachaitanyakc made their first contribution in #13145
• @huemin-art made their first contribution in #13314

Full Changelog: v0.18.1...v0.19.0

v0.18.2

Full Changelog: v0.18.1...v0.18.2

v0.18.1

What's Changed

• Fix canny node not working with fp16. by @comfyanonymous in #13085
• Fix sampling issue with fp16 intermediates. by @comfyanonymous in #13099
• Fix fp16 intermediates giving different results. by @comfyanonymous in #13100
• wan: vae: Fix light/color issue by @rattus128 in #13101

Full Changelog: v0.18.0...v0.18.1

v0.18.0

What's Changed

• Revert "Revert "feat: Add CacheProvider API for external distributed caching"" by @deepme987 in #12915
• fix(api-nodes): Tencent TextToModel and ImageToModel nodes by @bigcat88 in #12680
• Bump comfyui-frontend-package to 1.41.19 by @comfy-pr-bot in #12923
• bump manager version to 4.1b4 by @ltdrdata in #12930
• comfy aimdo 0.2.11 + Improved RAM Pressure release strategies - Windows speedups by @rattus128 in #12925
• Update README. by @comfyanonymous in #12931
• comfy-aimdo 0.2.12 by @rattus128 in #12941
• fix: use no-store cache headers to prevent stale frontend chunks by @christian-byrne in #12911
• feat: Support mxfp8 by @kijai in #12907
• Add --fp16-intermediates to use fp16 for intermediate values between … by @comfyanonymous in #12953
• Update comfyui-frontend-package to version 1.41.20 by @comfyanonymous in #12954
• LTXV: Accumulate VAE decode results on intermediate_device by @kijai in #12955
• bump manager version to 4.1b5 by @ltdrdata in #12957
• ops: opt out of deferred weight init if subclassed by @rattus128 in #12967
• Make EmptyLatentImage follow intermediate dtype. by @comfyanonymous in #12974
• Enable Pytorch Attention for AMD gfx1150 (Strix Point) by @lostdisc in #12973
• feat: add essentials_category to nodes and blueprints for Essentials tab by @christian-byrne in #12573
• feat(assets): align local API with cloud spec by @luke-mino-altherr in #12863
• ci: add check to block AI agent Co-authored-by trailers in PRs by @christian-byrne in #12799
• Skip running model finalizers at exit by @blepping in #12994
• Add --enable-dynamic-vram options to force enable it. by @comfyanonymous in #13002
• [Trainer] FP4, 8, 16 training by native dtype support and quant linear autograd function by @KohakuBlueleaf in #12681
• Fix potential issue. by @comfyanonymous in #13009
• fix: atomic writes for userdata to prevent data loss on crash by @christian-byrne in #12987
• Disable SageAttention for Hunyuan3D v2.1 DiT by @paulomuggler in #12772
• Update workflow templates to v0.9.26 by @comfyui-wiki in #13012
• Mark weight_dtype as advanced input in Load Diffusion Model node by @christian-byrne in #12769
• Reduce LTX VAE VRAM usage and save use cases from OOMs/Tiler by @rattus128 in #13013
• Reduce WAN VAE VRAM, Save use cases for OOM/Tiler by @rattus128 in #13014
• bump manager version to 4.1b6 by @ltdrdata in #13022
• Inplace VAE output processing to reduce peak RAM consumption. by @kijai in #13028
• Fix case where pixel space VAE could cause issues. by @comfyanonymous in #13030
• cascade: remove dead weight init code by @rattus128 in #13026
• fix: run text encoders on MPS GPU instead of CPU for Apple Silicon by @k06a in #12809
• fix(api-nodes): add support for "thought_image" in Nano Banana 2 by @bigcat88 in #13038
• Update comfyui-frontend-package version to 1.41.21 by @DrJKL in #13035
• Make more intermediate values follow the intermediate dtype. by @comfyanonymous in #13051
• Further Reduce LTX VAE decode peak RAM usage by @kijai in #13052
• Fix regression. by @comfyanonymous in #13053
• fp16 intermediates doen't work for some text enc models. by @comfyanonymous in #13056
• ltx: vae: implement chunked encoder + CPU IO chunking (Big VRAM reductions) by @rattus128 in #13062
• memory: Add more exclusion criteria to pinned read (fixes corrupt outputs - rare cases) by @rattus128 in #13067
• Reduce tiled decode peak memory by @kijai in #13050
• Revert "fix: run text encoders on MPS GPU instead of CPU for Apple Silicon" by @comfyanonymous in #13070
• Fix VRAM leak in tiler fallback in video VAEs by @rattus128 in #13073
• ltx: vae: Fix missing init variable by @rattus128 in #13074
• [API Nodes] mark seedream-3-0-t2i and seedance-1-0-lite models as deprecated by @bigcat88 in #13060
• Add slice_cond and per-model context window cond resizing by @drozbay in #12645
• feat(api-nodes): add Quiver SVG nodes by @bigcat88 in #13047
• Make EmptyImage node follow intermediate device/dtype. by @comfyanonymous in #13079
• Move inline comfy.context_windows imports to top-level in model_base.py by @Kosinkadink in #13083

New Contributors

• @lostdisc made their first contribution in #12973
• @paulomuggler made their first contribution in #12772
• @k06a made their first contribution in #12809

Full Changelog: v0.17.0...v0.18.0

v0.17.2

Full Changelog: v0.17.1...v0.17.2

v0.17.1

Full Changelog: v0.17.0...v0.17.1