Katsuhide Tsuruta
d12e618af4
- Call model.decoder() directly to bypass diffusers auto-scaling - Fix AlphaVAELoader directory scanning to skip non-directory entries - Remove incorrect latent scaling in encode/decode nodes
ComfyUI-AlphaVAE
ComfyUI custom nodes for AlphaVAE — native RGBA transparent image generation using FLUX.
AlphaVAE replaces the standard FLUX VAE with an RGBA-capable VAE, enabling direct transparent image generation without post-processing background removal.
Status
🚧 Work in Progress — Not yet functional.
Planned Nodes
| Node | Description |
|---|---|
AlphaVAE Loader |
Load AlphaVAE model (diffusers format) |
AlphaVAE Decode |
Decode latent to RGBA image (outputs IMAGE + MASK) |
AlphaVAE Encode |
Encode RGBA image to latent |
Requirements
- ComfyUI v0.18+
- FLUX.1-dev base model
- AlphaVAE model weights
License
Code: MIT License
Model License Notice
AlphaVAE model weights are a fine-tuned derivative of FLUX.1-dev's VAE. FLUX.1-dev is released under the FLUX.1 [dev] Non-Commercial License. As a derivative work, AlphaVAE weights may inherit FLUX.1-dev's non-commercial restrictions regardless of the license stated on the AlphaVAE repository. Commercial use may require a separate license from Black Forest Labs.
This project distributes only wrapper code, not model weights. Users are responsible for obtaining model weights and complying with applicable licenses.
References
- AlphaVAE Paper (arXiv:2507.09308)
- AlphaVAE GitHub
- AlphaVAE HuggingFace
- LayerDiffuse — Prior art (SDXL, unmaintained)