There are two different workflows for ComfyUI – one is a basic version, and the other is an optimized high-res version using Nunchaku.
Put together some custom ComfyUI nodes today that let you generate transparent images with Flux.
The idea sparked after exploring RedAIGC’s Flux-version-LayerDiffuse and the groundwork from lllyasviel’s LayerDiffuse (SD 1.5)—couldn’t resist building on top of their brilliant work.
Huge thanks to Milan Kastenmüller for all the support, feedback, and inspiration along the way.
The toolkit brings several nodes for ComfyUI, including:
TransparentVAE Loader for transparent images, configurable and compatible with standard VAE input.
Flux Decoder: Decodes latents into genuine RGBA transparent images, with augmentation for higher quality.
Save & Preview: Stores transparent PNGs (w/metadata) and previews them directly in ComfyUI with multiple background options.
If you’re interested in generating transparent images in ComfyUI workflows and want robust support for transparency, alpha channels, and advanced conditioning, you can check out the project and documentation here: