What's Changed

• Update nodes_model_patch.py by @comfyanonymous in #9597

Full Changelog: v0.3.53...v0.3.54

What's Changed

• Fix EasyCache/LazyCache crash when tensor shape/dtype/device changes during sampling by @blepping in #9528
• Remove models from readme that are not fully implemented. by @comfyanonymous in #9535
• Add models/audio_encoders directory. by @comfyanonymous in #9548
• Implement wav2vec2 as an audio encoder model. by @comfyanonymous in #9549
• Make AudioEncoderOutput usable in v3 node schema. by @comfyanonymous in #9554
• Update template to 0.1.66 by @comfyui-wiki in #9557
• Adding Google Gemini Image API node by @Kosinkadink in #9566
• Update template to 0.1.68 by @comfyui-wiki in #9569
• WIP Wan 2.2 S2V model. by @comfyanonymous in #9568
• Negative audio in s2v should be zeros. by @comfyanonymous in #9578
• Fix #9537 by @KohakuBlueleaf in #9576
• Fixes to make controlnet type models work on qwen edit and kontext. by @comfyanonymous in #9581
• Improve s2v performance when generating videos longer than 120 frames. by @comfyanonymous in #9582
• Better s2v memory estimation. by @comfyanonymous in #9584
• Add DPM++ 2M SDE Heun (RES) sampler by @Balladie in #9542
• Add a LatentConcat node. by @comfyanonymous in #9587
• Adjust flux mem usage factor a bit. by @comfyanonymous in #9588

New Contributors

• @Balladie made their first contribution in #9542

Full Changelog: v0.3.52...v0.3.53

What's Changed

• LTXV: fix key frame noise mask dimensions for when real noise mask exists by @harelc in #9425
• Add that qwen edit model is supported to readme. by @comfyanonymous in #9463
• Support for Qwen Diffsynth Controlnets canny and depth. by @comfyanonymous in #9465
• Fix: Handle invalid filenames during workflow save (#9434) by @saurabh-pingale in #9445
• Forgot this. by @comfyanonymous in #9470
• Support diffsynth inpaint controlnet/model patch. by @comfyanonymous in #9471
• convert String nodes to V3 schema by @bigcat88 in #9370
• v3 nodes (part a) by @bigcat88 in #9149
• [V3] convert Google Veo API node to the V3 schema by @bigcat88 in #9272
• [V3] convert Ideogram API nodes to the V3 schema by @bigcat88 in #9278
• Make it easier to implement future qwen controlnets. by @comfyanonymous in #9485
• Support InstantX Qwen controlnet. by @comfyanonymous in #9488
• feat(api-nodes): change "OpenAI Chat" display name to "OpenAI ChatGPT" by @bigcat88 in #9443
• feat(api-nodes): add copy button to Gemini Chat node by @bigcat88 in #9440
• Update template to 0.1.65 by @comfyui-wiki in #9501
• Add elementwise fusions by @contentis in #9495
• Implement EasyCache and Invent LazyCache by @Kosinkadink in #9496
• Fix Conditioning masks on 3d latents. by @comfyanonymous in #9506
• Python 3.13 is well supported. by @comfyanonymous in #9511
• Don't use the annoying new navigation mode by default. by @comfyanonymous in #9518
• Update frontend to v1.25.10 and revert navigation mode override by @christian-byrne in #9522

New Contributors

• @saurabh-pingale made their first contribution in #9445

Full Changelog: v0.3.51...v0.3.52

What's Changed

• Put back frontend version. by @comfyanonymous in #9317
• Reduce portable size again. by @comfyanonymous in #9323
• SDPA backend priority by @contentis in #9299
• Make last PR not crash comfy on old pytorch. by @comfyanonymous in #9324
• Fix XPU iGPU regressions by @simonlui in #9322
• Added context window support to core sampling code by @Kosinkadink in #9238
• av is an essential dependency. by @comfyanonymous in #9341
• Make custom node testing checkbox optional in issue templates by @webfiltered in #9342
• Update CODEOWNERS by @yoland68 in #9343
• Update default parameters for Moonvalley video nodes by @guill in #9290
• convert WAN nodes to V3 schema by @bigcat88 in #9201
• Make SLG nodes work on Qwen Image model. by @comfyanonymous in #9345
• Avoid torch compile graphbreak for older pytorch versions by @StrongerXi in #9344
• Fix last commit not working on older pytorch. by @comfyanonymous in #9346
• Add warning when using old pytorch. by @comfyanonymous in #9347
• fix(OpenAIGPTImage1): set correct MIME type for multipart uploads by @bigcat88 in #9348
• Add FluxKontextMultiReferenceLatentMethod node. by @comfyanonymous in #9356
• Implement wan2.2 camera model. by @comfyanonymous in #9357
• Update frontend to v1.25.8 by @christian-byrne in #9361
• record audio node by @jtydhr88 in #8716
• Qwen image model refactor. by @comfyanonymous in #9375
• WIP Qwen edit model: The diffusion model part. by @comfyanonymous in #9383
• Update template to 0.1.60 by @comfyui-wiki in #9377
• Make context step index detection much more robust by @Kosinkadink in #9392
• Bump frontend to 1.25.9 by @christian-byrne in #9394
• fix(WAN-nodes): invalid nodeid for WanTrackToVideo by @bigcat88 in #9396
• P2 of qwen edit model. by @comfyanonymous in #9412
• api_nodes: add kling-v2-1 and v2-1-master by @bigcat88 in #9257
• api_nodes: add MinimaxHailuoVideoNode node by @bigcat88 in #9262
• api_nodes: add GPT-5 series models by @bigcat88 in #9325
• api_nodes: release version of google's models by @bigcat88 in #9304
• feat(api-nodes): add Vidu Video nodes by @bigcat88 in #9368
• Change the TextEncodeQwenImageEdit node to use logic closer to refererence code. by @comfyanonymous in #9432
• Rope fix for qwen vl. by @comfyanonymous in #9435
• Qwen rotary embeddings should now match reference code. by @comfyanonymous in #9437
• Disable prompt weights for qwen. by @comfyanonymous in #9438
• Fix bf16 precision issue with qwen image embeddings. by @comfyanonymous in #9441
• Bump template to 0.1.62 by @comfyui-wiki in #9419

New Contributors

• @StrongerXi made their first contribution in #9344

Full Changelog: v0.3.50...v0.3.51

What's Changed

• Add Qwen Image model to readme. by @comfyanonymous in #9191
• Qwen Image model merging node. by @comfyanonymous in #9202
• qwenLora cannot load properly by @flybirdxx in #9208
• Update template to 0.1.52 by @comfyui-wiki in #9206
• Update frontend to v1.24.4 by @christian-byrne in #9175
• Fix RepeatLatentBatch not working on multi dim latents. by @comfyanonymous in #9227
• _ui.py import torchaudio safety check by @Kosinkadink in #9234
• async API nodes by @bigcat88 in #9129
• Users report gfx1201 is buggy on flux with pytorch attention. by @comfyanonymous in #9244
• Not sure if AMD actually support fp16 acc but it doesn't crash. by @comfyanonymous in #9258
• Bump pytorch cuda and rocm versions in readme instructions. by @comfyanonymous in #9273
• Only show feature flags log when verbose. by @comfyanonymous in #9281
• remove creation of non-used asyncio_loop by @bigcat88 in #9284
• Update template & embedded docs by @comfyui-wiki in #9283
• Support SimpleTuner lycoris lora for Qwen-Image by @PsychoLogicAu in #9280
• fix(Kling Image API Node): do not pass "image_type" when no image by @bigcat88 in #9271
• Update template to 0.1.58 by @comfyui-wiki in #9302
• Update test release package workflow with python 3.13 cu129. by @comfyanonymous in #9306
• Wan2.2 fun control support. by @comfyanonymous in #9292
• Make torchaudio exception catching less specific by @Kosinkadink in #9309
• Update template to 0.1.59 by @comfyui-wiki in #9313
• Update release workflow to python3.13 pytorch cu129 by @comfyanonymous in #9315
• Downgrade frontend for release. by @comfyanonymous in #9316

New Contributors

• @flybirdxx made their first contribution in #9208

Full Changelog: v0.3.49...v0.3.50

What's Changed

• API for Recently Used Items by @Myestery in #8792
• Update template to 0.1.47 by @comfyui-wiki in #9153
• Make sure all the conds are on the right device. by @comfyanonymous in #9151
• Make sure context is on the right device. by @comfyanonymous in #9154
• Cleanup. by @comfyanonymous in #9160
• Lower cond vram use by casting at the same time as device transfer. by @comfyanonymous in #9159
• Fix broken controlnet from last PR. by @comfyanonymous in #9167
• Various fixes for broken things from earlier PR. by @comfyanonymous in #9168
• Add some warnings and prevent crash when cond devices don't match. by @comfyanonymous in #9169
• Fix another issue with the PR. by @comfyanonymous in #9170
• Initial support for qwen image model. by @comfyanonymous in #9179
• Add Veo3 video generation node with audio support by @christian-byrne in #9110
• Update template to 0.1.48 by @comfyui-wiki in #9182
• Fix default shift and any latent size for qwen image model. by @comfyanonymous in #9186
• Update template to 0.1.51 by @comfyui-wiki in #9187

New Contributors

• @Myestery made their first contribution in #8792

Full Changelog: v0.3.48...v0.3.49

What's Changed

• ComfyAPI Core v0.0.2 by @guill in #8962
• Update template to 0.1.44 by @comfyui-wiki in #9104
• Small wan camera memory optimization. by @comfyanonymous in #9111
• Replace manual x * sigmoid(x) with torch silu in VAE nonlinearity by @chaObserv in #9057
• Add support for partial execution in backend by @guill in #9123
• More accurate explanation of release process. by @comfyanonymous in #9126
• V3 Node Schema Definition - initial by @Kosinkadink in #8656
• Fix WanFirstLastFrameToVideo node when no clip vision. by @comfyanonymous in #9134
• update template to 0.1.45 by @comfyui-wiki in #9135
• Tiny wan vae optimizations. by @comfyanonymous in #9136

Full Changelog: v0.3.47...v0.3.48

What's Changed

• Remove unecessary clones in the wan2.2 VAE. by @comfyanonymous in #9083
• Extra reserved vram on large cards on windows. by @comfyanonymous in #9093
• Make wan2.2 5B i2v take a lot less memory. by @comfyanonymous in #9102

Full Changelog: v0.3.46...v0.3.47

What's Changed

• Replace torchaudio.load with pyav. by @comfyanonymous in #8989
• Add hidream e1.1 example to readme. by @comfyanonymous in #8990
• This is not needed. by @comfyanonymous in #8991
• Try to fix line endings workflow. by @comfyanonymous in #9001
• Add xpu version for async offload and some other things. by @simonlui in #9004
• Add default device argument. by @comfyanonymous in #9023
• Fix xpu function not implemented. by @comfyanonymous in #9026
• Fix xpu function not implemented p2. by @comfyanonymous in #9027
• Enable pytorch attention by default for gfx1201 on torch 2.8 by @comfyanonymous in #9029
• Only enable cuda malloc on cuda torch. by @comfyanonymous in #9031
• Separate denoised and noise estimation in Euler CFG++ by @chaObserv in #9008
• [Training Node] algo support, grad acc, optional grad ckpt by @KohakuBlueleaf in #9015
• Support Iluvatar CoreX by @honglyua-il in #8585
• Added parameter required_frontend_version in the /system_stats API response by @shivansh-gupta4 in #8875
• Print xpu device name. by @comfyanonymous in #9035
• Add WAN ATI support by @ecfairle in #8874
• Remove useless code. by @comfyanonymous in #9041
• Update template to 0.1.40 by @comfyui-wiki in #9048
• Fix issue with line endings github workflow. by @comfyanonymous in #9053
• Update negative prompt for Moonvalley nodes by @Thor-ATX in #9038
• Add map_function to get_history. by @comfyanonymous in #9056
• Remove useless code. by @comfyanonymous in #9059
• Fix link from docs (due to document updates) by @comfyui-wiki in #9060
• Small readme update. by @comfyanonymous in #9071
• Add Moonvalley Marey V2V node with updated input validation by @christian-byrne in #9069
• Update template to 0.1.41 by @comfyui-wiki in #9079
• Wan 2.2 support. by @comfyanonymous in #9080
• Add wan2.2 to readme. by @comfyanonymous in #9081

New Contributors

• @honglyua-il made their first contribution in #8585
• @shivansh-gupta4 made their first contribution in #8875
• @ecfairle made their first contribution in #8874
• @Thor-ATX made their first contribution in #9038

Full Changelog: v0.3.45...v0.3.46

What's Changed

• Un-hardcode Chroma patch_size by @josephrocca in #8840
• Add reconstructed SA-Solver sampler by @chaObserv in #8834
• Support Multi Image-Caption dataset in lora training node by @KohakuBlueleaf in #8819
• Remove unmaintained notebook. by @comfyanonymous in #8845
• Better training loop implementation by @KohakuBlueleaf in #8820
• Add warning to catch torch import mistakes. by @comfyanonymous in #8852
• Support for async node functions by @guill in #8830
• Fix python3.9 by @comfyanonymous in #8858
• [fix] increase Kling API polling timeout to prevent user timeouts by @christian-byrne in #8860
• Remove windows line endings. by @comfyanonymous in #8866
• Execute async node earlier by @JettHu in #8865
• Fix fresca's input and output by @chaObserv in #8871
• Add model detection error hint for lora. by @comfyanonymous in #8880
• Remove tokens from history items by @christian-byrne in #8889
• Refactor previous pr. by @comfyanonymous in #8893
• Remove windows line endings. by @comfyanonymous in #8902
• Upate template to 0.1.36 by @comfyui-wiki in #8904
• Allow the prompt request to specify the prompt ID. by @FeepingCreature in #8189
• Fix wrong reference bug by @yoland68 in #8910
• Update nodes_gemini.py by @yoland68 in #8912
• Add action to detect windows line endings. by @comfyanonymous in #8917
• Fix MaskComposite error when destination has 2 dimensions by @bman654 in #8915
• Add attempt to work around the safetensors mmap issue. by @comfyanonymous in #8928
• LTXV VAE decoder: switch default padding mode by @harelc in #8930
• Silence clip tokenizer warning. by @comfyanonymous in #8934
• Experimental CFGNorm node. by @comfyanonymous in #8942
• Forgot this by @comfyanonymous in #8957
• Add nested style of dual cfg to DualCFGGuider node. by @comfyanonymous in #8965
• Update template to 0.1.37 by @comfyui-wiki in #8967
• Add SamplingPercentToSigma node by @chaObserv in #8963
• Document what the fast_fp16_accumulation is in the portable. by @comfyanonymous in #8973
• Update template to 0.1.39 by @comfyui-wiki in #8981

New Contributors

• @josephrocca made their first contribution in #8840
• @bman654 made their first contribution in #8915

Full Changelog: v0.3.44...v0.3.45