ComfyUI v0.38.0
retained text
# Links * [**Official changelog**](https://docs.comfy.org/changelog#v0-38-0) * [**GitHub release**](https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.38.0) # New Open-Source Model Support * [**Ming Image 0.1 Design**](https://github.com/Comfy-Org/ComfyUI/pull/16482): A 6B text-to-image model for text-heavy design layouts, with transparent RGBA output * [**MiniMax-H3 Fun ControlNet Union 2.0**](https://github.com/Comfy-Org/ComfyUI/pull/16471): Combine multiple control conditions in one H3 generation * [**ID-V2V**](https://github.com/Comfy-Org/ComfyUI/pull/15139): Identity-preserving video-to-video on Wan 2.1 and VACE based models * [**Qwen-Image 2.1 ControlNet**](https://github.com/Comfy-Org/ComfyUI/pull/16519): Union Fun ControlNet support for Qwen-Image 2.1 editing * [**Qwen-Image 2.1 tiny VAE**](https://github.com/Comfy-Org/ComfyUI/pull/16552): A tiny VAE for faster previews and lower memory use # New Node Updates * [**Convert Image Color Space**](https://github.com/Comfy-Org/ComfyUI/pull/16541): Added the LogC3 and ACEScct HDR color spaces * [**w6a8 quantization**](https://github.com/Comfy-Org/ComfyUI/pull/16483): Added support for the w6a8 format * [**JsonExtractString**](https://github.com/Comfy-Org/ComfyUI/pull/16439): Parses JSON that is wrapped in a prefix or a suffix * [**Per-block attention**](https://github.com/Comfy-Org/ComfyUI/pull/16419): Model files can declare which attention implementation each block uses # Partner Node Updates * [**Seedance 2.5 Draft mode**](https://github.com/Comfy-Org/ComfyUI/pull/16529): Renders a fast 480p preview whose `draft_task_id` drives the 1080p final render * [**Seedream 5.0 Flash**](https://github.com/Comfy-Org/ComfyUI/pull/16522): A faster, lower-cost model on the Seedream node, and the Pro max resolution now reaches 4.62MP * [**GPT-6 Sol and Luna**](https://github.com/Comfy-Org/ComfyUI/pull/16520): Two lower-cost GPT-6 tiers on the OpenAI Chat node * [**Sora nodes removed**](https://github.com/Comfy-Org/ComfyUI/pull/16609): The retired Sora video nodes are gone from ComfyUI # Performance & Stability * [**Async weight offload on NPU**](https://github.com/Comfy-Org/ComfyUI/pull/16057): The async weight offload path now runs on Ascend NPU devices * [**ACE-Step 1.5**](https://github.com/Comfy-Org/ComfyUI/pull/16461): CUDA graphs and the memory compiler on the autoregressive model * [**Yue2 AR**](https://github.com/Comfy-Org/ComfyUI/pull/16626): Faster autoregressive generation for Yue2 * [**SeedVR2**](https://github.com/Comfy-Org/ComfyUI/pull/16530): Optimized video upscaling * [**Fast disk detection**](https://github.com/Comfy-Org/ComfyUI/pull/16425): Extended to all model loaders * [**Asset scans**](https://github.com/Comfy-Org/ComfyUI/pull/16546): Asset scans and output rescans no longer block the UI on large libraries * [**Lumina models**](https://github.com/Comfy-Org/ComfyUI/pull/16515): Lower memory usage and `comfy_attention` support * [**torchaudio removed**](https://github.com/Comfy-Org/ComfyUI/pull/16457): Dropped the torchaudio dependency # Bug Fixes * [**ImageUpscaleWithModel**](https://github.com/Comfy-Org/ComfyUI/pull/16500): No longer crashes on RGBA images, and transparency survives the upscale * [**Qwen VL preprocessing**](https://github.com/Comfy-Org/ComfyUI/pull/16548): Fixed a crash on 4-channel RGBA reference images * [**MiniMax-H3 VAE**](https://github.com/Comfy-Org/ComfyUI/pull/16485): Fixed an `rms_rope` crash on offloaded `qk_norm_scale` * [**HDR output**](https://github.com/Comfy-Org/ComfyUI/pull/16565): Fixed HDR output with negative values in Convert Image Color Space # Workflow Templates * [**Templates v0.11.70**](https://github.com/Comfy-Org/ComfyUI/pull/16557): Updated the bundled workflow templates
document fetched 2026-10-01 00:45 · canonical https://reddit.com/r/comfyui/comments/1wtxujx/comfyui_v0380