• Fix Tiled VAE decoding on LTX-Video
• Fix some lowvram mode issues with LTX-Video

Full Changelog: v0.3.3...v0.3.4

• Fix issues with the lowvram system.

Full Changelog: v0.3.2...v0.3.3

• Support LTX-Video model

Full Changelog: v0.3.1...v0.3.2

• Support new "Flux Tools" models: Blog Post

Full Changelog: v0.3.0...v0.3.1

What's Changed

• New SkipLayerGuidanceDiT node that supports most DiT models (Flux, Auraflow, Stable Audio, SD3, SD3.5, Mochi).
• Fast Mochi previews.
• Allow POST /userdata/{file} endpoint to return full file info by @huchenlei in #5446
• fix dynamic shape export by @contentis in #5490
• arbitrary downscale fix for SAG with Kohya Deep Shrink by @DenOfEquity in #5546
• Live terminal output by @pythongosssss in #5396
• improve: add descriptions for clip loaders by @ltdrdata in #5576
• fix --cuda-device arg for AMD/HIP devices by @Bratzmeister in #5586
• Update web content to release v1.3.44 by @huchenlei in #5620
• Fix 3.8 compatibility in user_manager.py by @huchenlei in #5645
• Update UI screenshot in README by @huchenlei in #5666
• Add terminal size fallback by @yoland68 in #5623

New Contributors

• @DenOfEquity made their first contribution in #5546
• @Bratzmeister made their first contribution in #5586

Full Changelog: v0.2.7...v0.3.0

What's Changed

• Mochi VAE encoder support.
• Support for SD3.5 medium loras in diffusers format.
• Rename ImageCrop to Image Crop by @KoreTeknology in #5424
• Avoid SyntaxWarning in UniPC docstring by @akx in #5442
• Rename some nodes in Display Name Mappings (nodes.py) by @KoreTeknology in #5439

New Contributors

• @KoreTeknology made their first contribution in #5424
• @akx made their first contribution in #5442

Full Changelog: v0.2.6...v0.2.7

What's Changed

• SLG first implementation for SD3.5 by @Dango233 in #5404
• Update web content to release v1.3.26 by @huchenlei in #5413

Full Changelog: v0.2.5...v0.2.6

What's Changed

• SD3.5 Medium.
• WIP Mochi support 5cbb01b
• Update web content to release v1.3.21 by @huchenlei in #5351
• Euler Ancestral now works on Flow models d605677
• remove attention abstraction in mmdit by @contentis in #5324
• support SimpleTuner lycoris lora for SD3 by @PsychoLogicAu in #5340
• Revert download to .tmp in frontend_management by @huchenlei in #5369

New Contributors

• @contentis made their first contribution in #5324
• @PsychoLogicAu made their first contribution in #5340

Full Changelog: v0.2.4...v0.2.5

What's Changed

• Update windows standalone package to python 3.12 and pytorch 2.5
• Support for checkpoints with scaled fp8 model weights for best quality when using the scaled fp8 matrix multiplication on hardware that supports it (Nvidia 40 series/ada/H100 and above).
• Some new experimental nodes to do latent operations (latent->advanced->operations).
• 🐛 Add device to variable c by @kadirnar in #5210
• Fix Transformers FutureWarning by @svdC1 in #5140
• Frontend Manager: avoid redundant gh calls for static versions by @mcmonkey4eva in #5152
• Ignore more network related errors during websocket communication. by @xiaodong-xie in #5269

New Contributors

• @kadirnar made their first contribution in #5210
• @svdC1 made their first contribution in #5140
• @xiaodong-xie made their first contribution in #5269

Full Changelog: v0.2.3...v0.2.4

What's Changed

• Update web content to release v1.3.11 by @huchenlei in #5189
• Add a weight_dtype fp8_e4m3fn_fast to the Diffusion Model Loader node.
• Fix flux doras with diffusers keys.
• Slightly improve fast previews for Flux, SDXL and SD3
• Let --verbose have an argument for the log level.
• Allow using SD3 type te output on flux model.
• torch.compile node (TorchCompileModel)
• Add an optional VAE input to the ControlNetApplyAdvanced node and deprecate the other controlnet nodes.
• Load controlnet in fp8 if weights are in fp8.
• Make --listen listen on both ipv4 and ipv6 at the same time by default.
• VAEDecodeAudio now does some normalization on the audio.
• Support AliMama SD3 and Flux inpaint controlnets (ControlNetInpaintingAliMamaApply).
• Prevent random websites from sending blind requests to local comfyui instance.
• Add a LoraSave node to save a lora from a ModelMergeSubtract and/or CLIPMergeSubtract output.
• Fix scipy.stats import when using beta scheduler on API by @nom in #4815
• Fix error with ExecutionBlocker and OUTPUT_IS_LIST by @guill in #4836
• fix: url decode filename from API by @w1gs in #4801
• allow current timestamp in save image prefix by @mcmonkey4eva in #4030
• Add .venv to gitignore by @david02871 in #4756
• Add back CI action for tests-ui by @huchenlei in #4859
• Expand user directory for basepath in extra_models_paths.yaml by @robinjhuang in #4857
• [🗻 Mount Fuji Commit] Add unit tests for folder path utilities by @yoland68 in #4869
• Add content-type filter method to folder_paths by @christian-byrne in #4054
• Add CFG++ to DPM++ 2S Ancestral by @yoinked-h in #3871
• Expand variables in base_path for extra_config_paths.yaml. by @robinjhuang in #4893
• Minor: update tests-unit README.md by @yoland68 in #4896
• Add cli arg to override user directory by @robinjhuang in #4856
• Reduce repeated calls of get_immediate_node_signature in cache by @JettHu in #4871
• Add full_info flag in /userdata endpoint to list out file size and last modified timestamp by @huchenlei in #4905
• Normalize path returned by /userdata to always use / as separator by @huchenlei in #4906
• Fix RecursionError for (very) large graphs by @Acly in #4886
• Remove empty spaces in user_manager.py by @huchenlei in #4917
• Reduce repeated calls of INPUT_TYPES in cache by @JettHu in #4922
• Improve tiling calculations to reduce number of tiles that need to be processed. by @pharmapsychotic in #4944
• add simple error check to model loading by @mcmonkey4eva in #4950
• Add route to list model types by @ImDarkTom in #4846
• very simple strong-cache on model list by @mcmonkey4eva in #4969
• Fix xpu Pytorch nightly build from calling ipex.optimize which doesn't exist. by @simonlui in #4978
• add 'is_default' to model paths config by @mcmonkey4eva in #4979
• add internal /folder_paths route by @mcmonkey4eva in #4980
• add laplace scheduler [2407.03297] by @yoinked-h in #4990
• Update web content to release v1.2.60 by @huchenlei in #5017
• Added to the websockets_api_example.py to show how to decode latent previews from the binary stream by @RandomGitUser321 in #5016
• Add ws.close() to the websocket examples by @RandomGitUser321 in #5020
• Run unit tests on Windows/MacOS as well. by @robinjhuang in #5018
• Add dpmpp_2m_cfg_pp sampler by @chaObserv in #4992
• Internal download API: Add proper validated directory input by @mcmonkey4eva in #4981
• Flux torch compile fix by @city96 in #5082
• IS_CHANGED should be a classmethod by @LexLuc in #5159
• Update web content to release v1.2.64 by @huchenlei in #5124
• Add a try/catch to skip special file_name values instead of failing to load a node. by @brendanhoar in #5187
• Change runner label for the new runners by @yoland68 in #5197
• Increase maximum macOS version to 15.0.1 when forcing upcast attention by @joneavila in #5191
• Hotfix for the div zero occurrence when memory_used_encode is 0 by @ltdrdata in #5121

New Contributors

• @nom made their first contribution in #4815
• @w1gs made their first contribution in #4801
• @david02871 made their first contribution in #4756
• @yoinked-h made their first contribution in #3871
• @Acly made their first contribution in #4886
• @pharmapsychotic made their first contribution in #4944
• @ImDarkTom made their first contribution in #4846
• @RandomGitUser321 made their first contribution in #5016
• @chaObserv made their first contribution in #4992
• @LexLuc made their first contribution in #5159
• @brendanhoar made their first contribution in #5187
• @joneavila made their first contribution in #5191

Full Changelog: v0.2.2...v0.2.3