ComfyUI v0.32.0 drops with 20+ fixes and new models
ComfyUI v0.32.0 introduces 20+ fixes, new model support, and improved debugging...
ComfyUI, the open-source AI image generation framework beloved by Stable Diffusion and ComfyUI users, has released v0.32.0 with over 20 fixes and improvements. The update raises the minimum officially supported PyTorch version to 2.7 and addresses critical issues like VAEDecodeTiled crashes on NestedTensor latents and memory spikes with MiniMax H3 models. Debugging gets a boost with better nested tensor handling and more visible CUDA warnings.
The release also expands creative tooling with new partner nodes: LTX 2.5 support arrives with dedicated nodes, while Qwen-Image 3.0 Pro and Grok-Imagine-Image-2.0 models can now be integrated via ComfyUI's modular system. Backend optimizations include Comfy Kitchen attention improvements and previewable outputs for Media Assets badges, alongside API enhancements like new tag filters for assets lists.
- ComfyUI v0.32.0 includes 20+ fixes, new model support (LTX 2.5, Qwen-Image 3.0 Pro, Grok-Imagine-Image-2.0)
- Raises minimum PyTorch support to 2.7 and improves debugging with nested tensor fixes
- Adds previewable outputs, new API filters, and optimizations like Comfy Kitchen attention
Why It Matters
Critical fixes and expanded model support make ComfyUI more stable and versatile for production AI image pipelines.