What's Changed

• feat: add Math Expression node with simpleeval evaluation by @dante01yoon in #12687
• feat(api-nodes): add TencentSmartTopology node by @bigcat88 in #12741
• feat(api-nodes): add Gemini 3.1 Flash-Lite model to LLM node by @bigcat88 in #12803
• Fix fp16 audio encoder models by @rattus128 in #12811
• Don't run TE on cpu when dynamic vram enabled. by @comfyanonymous in #12815
• Fix requirements version. by @comfyanonymous in #12817
• Update workflow templates to v0.9.11 by @comfyui-wiki in #12821
• mm: reset_cast_buffers: sync compute stream before free by @rattus128 in #12822

New Contributors

• @dante01yoon made their first contribution in #12687

Full Changelog: v0.16.3...v0.16.4

What's Changed

• Add manual cast to LTX2 vocoder conv_transpose1d by @kijai in #12795
• LTX audio vae novram fixes. by @comfyanonymous in #12796
• feat: add causal_fix parameter to add_keyframe_index and append_keyframe by @tavihalperin in #12797

Full Changelog: v0.16.2...v0.16.3

What's Changed

• comfy-aimdo 0.2.7 by @rattus128 in #12791
• ops: Handle CPU weight in VBAR caster by @rattus128 in #12792
• Update workflow templates to v0.9.10 by @comfyui-wiki in #12793
• Refactor. by @comfyanonymous in #12794

Full Changelog: v0.16.1...v0.16.2

What's Changed

• feat(api-nodes-xAI): updated models and pricing by @bigcat88 in #12756
• feat(api-nodes): enable Kling 3.0 Motion Control by @bigcat88 in #12785
• Update workflow templates to v0.9.8 by @comfyui-wiki in #12788

Full Changelog: v0.16.0...v0.16.1

What's Changed

• Portable with cu128 isn't useful anymore. by @comfyanonymous in #12666
• feat(ace-step): add ACE-Step 1.5 lycoris key alias mapping for LoKR #… by @fappaz in #12665
• bug_report template: Push harder for logs by @rattus128 in #12657
• feat: Support SDPose-OOD by @kijai in #12661
• Enable Pytorch Attention for gfx950 by @vickytsang in #12641
• Fix GLSL node requiring all input images to have same dimensions by @pythongosssss in #12673
• feat: add ResolutionSelector node for aspect ratio presets by @christian-byrne in #12199
• Class WanVAE, def encode, feat_map is using self.decoder instead of self.encoder by @ReinerBforartists in #12682
• ops: implement lora requanting for non QuantizedTensor fp8 by @rattus128 in #12668
• Limit overlap in image tile and combine nodes to prevent issues. by @comfyanonymous in #12688
• Native LongCat-Image implementation by @Talmaj in #12597
• PyOpenGL-accelerate is not necessary. by @comfyanonymous in #12692
• refactor: use enum members as ASPECT_RATIOS dict keys by @christian-byrne in #12689
• fix(ace15): handle missing lm_metadata in memory estimation during ch… by @fappaz in #12686
• feat: Support SCAIL WanVideo model by @kijai in #12614
• Disable dynamic_vram when weight hooks applied by @rattus128 in #12653
• cli_args: Default comfy to DynamicVram mode by @rattus128 in #12658
• Move parsing of requirements logic to function. by @comfyanonymous in #12701
• Disable dynamic vram on wsl. by @comfyanonymous in #12706
• refactor: rename Mahiro CFG to Similarity-Adaptive Guidance by @christian-byrne in #12172
• feat: add text preview support to jobs API by @christian-byrne in #12169
• fix: handle substep sigmas in context window set_step by @drozbay in #12719
• comfy aimdo 0.2.3 by @rattus128 in #12720
• Make --disable-smart-memory disable dynamic vram. by @comfyanonymous in #12722
• model_mangament: make dynamic --disable-smart-memory work by @rattus128 in #12724
• comfy aimdo 0.2.4 - Fix unaligned allocator sizes by @rattus128 in #12727
• feat(api-nodes): allow to use "IMAGE+TEXT" in NanoBanana2 by @bigcat88 in #12729
• Update workflow templates to v0.9.5 by @comfyui-wiki in #12732
• Support loading zeta chroma weights properly. by @comfyanonymous in #12734
• Feat: z-image pixel space (model still training atm) by @lodestone-rock in #12709
• Fix VAEDecodeAudioTiled ignoring tile_size input (#12735) by @xeinherjer-dev in #12738
• ModelPatcherDynamic: Force load all non-comfy weights by @rattus128 in #12739
• main: Load aimdo after logger is setup by @rattus128 in #12743
• Adjust memory usage factor of zeta model. by @comfyanonymous in #12746
• CURVE type by @jtydhr88 in #12581
• DynamicVram: Add casting / fix torch Buffer weights by @rattus128 in #12749
• Fix VideoFromComponents.save_to crash when writing to BytesIO by @artokun in #12683
• model_patcher: Improve dynamic offload heuristic by @rattus128 in #12759
• Comfy Aimdo 0.2.5 + Fix offload performance in DynamicVram by @rattus128 in #12754
• Pass tokens when loading text gen model for text generation by @kijai in #12755
• comfy-aimdo 0.2.6 by @rattus128 in #12764
• Initial load device should be cpu when using dynamic vram. by @comfyanonymous in #12766
• Support the LTXAV 2.3 model. by @comfyanonymous in #12773
• Fix cublas ops on dynamic vram. by @comfyanonymous in #12776
• Update workflow templates to v0.9.6 by @comfyui-wiki in #12778
• Fix cuda getting initialized in cpu mode. by @comfyanonymous in #12779
• Update workflow templates to v0.9.7 by @comfyui-wiki in #12780

New Contributors

• @fappaz made their first contribution in #12665
• @vickytsang made their first contribution in #12641
• @ReinerBforartists made their first contribution in #12682
• @Talmaj made their first contribution in #12597
• @xeinherjer-dev made their first contribution in #12738

Full Changelog: v0.15.1...v0.16.0

What's Changed

• Fix KeyError when prompt entries lack class_type key by @Kosinkadink in #12595
• comfy aimdo 0.2.1 by @rattus128 in #12620
• Disable dynamic_vram when using torch compiler by @rattus128 in #12612
• Fix Aimdo fallback on probe to not use zero-copy sft by @rattus128 in #12634
• comfy aimdo 0.2.2 by @rattus128 in #12635
• Patch frontend to 1.39.19 (from 1.39.16) (fix progress text serialization in API nodes) by @christian-byrne in #12642
• Fix LTXAV text enc min length. by @comfyanonymous in #12640
• Fix ltxav te mem estimation. by @comfyanonymous in #12643
• feat: Support WanVideo based segmentation model FlowRVS by @kijai in #12637
• feat: per-guide attention strength control in self-attention by @tavihalperin in #12518
• Cleanups to the last PR. by @comfyanonymous in #12646
• fix: move essentials_category to correct replacement nodes by @christian-byrne in #12568
• Fix GLSL node with pyopengl accelerate numpy textures by @pythongosssss in #12650
• main: load aimdo earlier by @rattus128 in #12655
• feat(api-nodes): add NanoBanana2 by @bigcat88 in #12660
• Update workflow templates to v0.9.4 by @comfyui-wiki in #12664

Full Changelog: v0.15.0...v0.15.1

What's Changed

• ops: limit return of requants (Fix perf of some fp8 dynamic_vram workflows) by @rattus128 in #12506
• Bump comfyui-frontend-package to 1.39.14 by @comfy-pr-bot in #12494
• BBox widget by @jtydhr88 in #11594
• fix: use glob matching for Gemini image MIME types by @huntcsg in #12511
• fix(api-nodes): add price badge for Rodin Gen-2 node by @bigcat88 in #12512
• Add simple 3 band equalizer node for audio. by @comfyanonymous in #12519
• feat: Add basic text generation support with native models, initially supporting Gemma3 by @kijai in #12392
• fix(api-nodes): force Gemini to return uncompressed images by @bigcat88 in #12516
• feat: add essentials_category by @Yourz in #12357
• feat: mark 429 widgets as advanced for collapsible UI by @christian-byrne in #12197
• Small cleanup and try to get qwen 3 work with the text gen. by @comfyanonymous in #12537
• Force min length 1 when tokenizing for text generation. by @comfyanonymous in #12538
• Add GLSL shader node using PyOpenGL by @pythongosssss in #12148
• [API Nodes] add ElevenLabs nodes by @bigcat88 in #12207
• chore: add CodeRabbit configuration for automated code review by @christian-byrne in #12539
• Update nightly installation command for ROCm by @comfyanonymous in #12547
• Fix non-contiguous audio waveform crash in video save by @artokun in #12550
• feat: add gradient-slider display mode for FLOAT inputs by @jtydhr88 in #12536
• update glsl blueprint with gradient by @jtydhr88 in #12548
• feat: add essential subgraph blueprints by @christian-byrne in #12552
• fix: swap essentials_category from CLIPTextEncode to PrimitiveStringM… by @christian-byrne in #12553
• Add support for pyopengl < 3.1.4 where the size parameter does not exist by @pythongosssss in #12555
• comfy-aimdo 0.2 - Improved pytorch allocator integration by @rattus128 in #12557
• fix: specify UTF-8 encoding when reading subgraph files by @DrJKL in #12563
• Add category to Normalized Attention Guidance node by @christian-byrne in #12565
• chore: tune CodeRabbit config to limit review scope and disable for drafts by @christian-byrne in #12567
• Move LTXAV av embedding connectors to diffusion model. by @comfyanonymous in #12569
• Fix dtype issue in embeddings connector. by @comfyanonymous in #12570
• Fix issue loading fp8 ltxav checkpoints. by @comfyanonymous in #12582
• feat(api-nodes): add KlingAvatar node by @bigcat88 in #12591
• Update workflow templates to v0.9.2 by @comfyui-wiki in #12596
• Add SplitImageToTileList and ImageMergeTileList nodes. by @comfyanonymous in #12599
• Temporality unbreak some LTXAV (LTX 2.0) workflows to give people time to migrate. by @comfyanonymous in #12605
• Patch frontend to 1.39.16 (from 1.39.14) by @christian-byrne in #12604
• feat(api-nodes): add ByteDance Seedream-5 model by @bigcat88 in #12609
• Update workflow templates to v0.9.3 by @comfyui-wiki in #12610
• Don't crash right away if op is uninitialized. by @comfyanonymous in #12615
• Update embedded docs to v0.4.3 by @comfyui-wiki in #12601

New Contributors

• @huntcsg made their first contribution in #12511
• @Yourz made their first contribution in #12357
• @artokun made their first contribution in #12550

Full Changelog: v0.14.1...v0.15.0

• Fix for Gemini/Nano banana api nodes that sometimes returned a blank image.
  Full Changelog: v0.14.1...v0.14.2

What's Changed

• Fix anima LLM adapter forward when manual cast by @chaObserv in #12504
• feat(api-nodes): add "viduq3-turbo" model support by @bigcat88 in #12482
• feat(api-nodes): add Recraft V4 nodes by @bigcat88 in #12502
• Update workflow templates to v0.8.43 by @comfyui-wiki in #12507

Full Changelog: v0.14.0...v0.14.1

What's Changed

• mp: dont deep-clone objects from model_options by @rattus128 in #12382
• sd: delay VAE dtype archive until after override by @rattus128 in #12388
• dynamic_vram: ops: Fix vanilla-fp8 loaded lora quality by @rattus128 in #12390
• Fix 3D models not showing in outputs window by @guill in #12381
• Add a VideoSlice node by @AustinMroz in #12107
• [Trainer] training with proper offloading by @KohakuBlueleaf in #12189
• Make built in lora training work on anima. by @comfyanonymous in #12402
• fix(api-nodes): retry on connection errors during polling instead of aborting by @bigcat88 in #12393
• Dispatch desktop auto-bump when a ComfyUI release is published by @benceruleanlu in #12398
• [API Nodes] enable Magnific Upscalers by @bigcat88 in #12179
• dynamic_vram: Fix windows Aimdo crash + Fix LLM performance by @rattus128 in #12408
• ace15: Use dynamic_vram friendly trange by @rattus128 in #12409
• model_patcher: guard against none model_dtype by @rattus128 in #12410
• Update ace15.py to allow min_p sampling by @askmyteapot in #12373
• Add a tip for common error. by @comfyanonymous in #12414
• fix(api-nodes): add separate retry budget for 429 rate limit responses by @bigcat88 in #12421
• ModelPatcherDynamic: force load non leaf weights by @rattus128 in #12433
• llama: use a more efficient rope implementation by @rattus128 in #12434
• Update command to install AMD stable linux pytorch. by @comfyanonymous in #12437
• dynamic_vram: Training fixes by @rattus128 in #12442
• Use torch RMSNorm for flux models and refactor hunyuan video code. by @comfyanonymous in #12432
• Support generating attention masks for left padded text encoders. by @comfyanonymous in #12454
• Fix some custom nodes. by @comfyanonymous in #12455
• Add left padding to LTXAV text encoder. by @comfyanonymous in #12456
• Add working Qwen 2512 ControlNet (Fun ControlNet) support by @krigeta in #12359
• Patch frontend to 1.38.14 (from 1.38.13) by @christian-byrne in #12469
• Remove unsafe pickle loading code that was used on pytorch older than 2.4 by @comfyanonymous in #12473
• chore(api-nodes): remove "gpt-4o" model by @bigcat88 in #12467
• Node Replacement API by @Kosinkadink in #12014
• feat(api-nodes-Tencent): add ModelTo3DUV, 3DTextureEdit, 3DParts nodes by @bigcat88 in #12428
• feat(api-nodes): add Bria RMBG nodes by @bigcat88 in #12465
• Fix lora Extraction in offload conditions (+ dynamic_vram mode) by @rattus128 in #12479
• MPDynamic: force load flux img_in weight (Fixes flux1 canny+depth lora crash) by @rattus128 in #12446
• Remove workaround for old pytorch. by @comfyanonymous in #12480
• Update workflow templates to v0.8.42 by @comfyui-wiki in #12491
• Remove code to support RMSNorm on old pytorch. by @comfyanonymous in #12499
• gitignore venv* by @alexheretic in #12431
• Allow control_after_generate to be type ControlAfterGenerate in v3 schema by @Kosinkadink in #12187
• Implement NAG on all the models based on the Flux code. by @comfyanonymous in #12500
• Fix anima preprocess text embeds not using right inference dtype. by @comfyanonymous in #12501

New Contributors

• @askmyteapot made their first contribution in #12373
• @krigeta made their first contribution in #12359

Full Changelog: v0.13.0...v0.14.0