Qwen Image Layered (ComfyUI weights)
This Hugging Face repo is a ComfyUI-oriented packaging of the Qwen Image Layered diffusion checkpoint. Instead of a single “mystery file,” it ships the core diffusion weights in multiple formats (including bf16 and an fp8-mixed variant) plus a matching VAE under split_files/, which makes it easier to wire into a node-based workflow without hunting for the right companion components.
If you’re already living in ComfyUI, this is the kind of release that’s worth bookmarking: it’s not so much a new model family as a practical delivery format. The fastest way to try it is to drop the diffusion weights and VAE into your ComfyUI models directories, load the checkpoint in a minimal text-to-image graph, and then iterate toward your real pipeline (hi-res fix, ControlNet, inpainting, or whatever you actually ship). If you run into quality/perf tradeoffs, compare bf16 vs. fp8mixed on the same prompt and seed to see whether fp8 is “good enough” for your use case.
Quick stats from the listing feed: pipeline: text-to-image · 22 likes.
Source listing: https://huggingface.co/models?sort=modified