v0.32.0

What's Changed

• Make it easier to debug nested tensors. by @comfyanonymous in #15383
• Update workflow templates to v0.11.37 by @comfyui-wiki in #15415
• Minimum officially supported pytorch is now 2.7 by @comfyanonymous in #15413
• fix: make Create Layered Image discoverable and its flags self-explanatory by @jtydhr88 in #15429
• Fix upscale models breaking on non dynamic vram low vram. by @comfyanonymous in #15437
• Add previewable_outputs_count to /api/jobs (backend half of Media Assets badge fix) by @claude[bot] in #15148
• Extend ER-SDE noise scaler by scaling h(t) by @chaObserv in #15428
• Make a context manager for cast_bias_weight and use it. by @blepping in #14750
• Optimize MiniMax-H3 VAE by @kijai in #15446
• [Partner Nodes] feat(Qwen): add Qwen-Image 3.0 Pro by @bigcat88 in #15327
• Add tags_all / tags_any / tags_none tag filters to the assets list API by @synap5e in #15332
• Fix VAEDecodeTiled crash on NestedTensor latents (MiniMax H3) by @chelsealong in #15477
• Make cu130 warning more visible. by @comfyanonymous in #15463
• Implement comfy kitchen attention. by @comfyanonymous in #15479
• Fix peak memory issue with H3. by @comfyanonymous in #15486
• Add support for LTX 2.5 by @alexisrolland in #15499
• [Partner Nodes] feat(LTX): add new nodes for model version 2.5 by @bigcat88 in #15501
• Fix for broken tiled audio decode. by @comfyanonymous in #15502
• Update workflow templates to v0.11.39 by @comfyui-wiki in #15504
• [Partner Nodes] feat(Grok): add Grok-Imagine-Image-2.0 model by @bigcat88 in #15496
• Fix some clip vision regression. by @comfyanonymous in #15506
• Don't depend on transformers for mistral and llama tokenizers. by @comfyanonymous in #15503
• Remove potentially problematic process_tokens method. by @comfyanonymous in #15507

New Contributors

• @claude[bot] made their first contribution in #15148
• @chelsealong made their first contribution in #15477

Full Changelog: v0.31.0...v0.32.0