Characters, illustrations, and little objects — ready for your composition.

**CKMoss-Image-RGBA** includes an RGBA VAE and an optional DiT LoRA, allowing MOSS-Image to output PNG files with an alpha channel for character assets, illustration, and layered compositions. **Images can be generated without the DiT LoRA.** Keep the original MOSS-Image model and load this RGBA VAE. The optional LoRA adapts MOSS-Image-SFT for transparent-image generation; background transparency and edge quality can differ between the two setups. ## Downloads | File | Purpose | Destination | | --- | --- | --- | | [RGBA VAE · 336 MB](vae/moss_rgba_vae.safetensors) | Required; already merged and ready to load | `ComfyUI/models/vae/` | | [DiT LoRA · 298 MB](loras/moss_rgba_dit_lora.safetensors) | Optional; use with MOSS-Image-SFT | `ComfyUI/models/loras/` | | [ComfyUI nodes on GitHub](https://github.com/Chenkin-x/ComfyUI-CKMoss-Image-RGBA) | Nodes, guides, and example workflows | Clone into `ComfyUI/custom_nodes/` | | [Model code on GitHub](https://github.com/Chenkin-x/CKMoss-Image-RGBA) | Python entry points and model code | Clone and run from the project root | This repository contains a **VAE and a LoRA**. Obtain the MOSS-Image base model and its text encoder separately. See the [official ComfyUI instructions](https://github.com/OpenMOSS/MOSS-Image/blob/main/comfyui/README.md). ## Get started Install the nodes, place the weights, and drag a workflow into ComfyUI. Select your base model and text encoder: - [RGBA VAE only](workflows/vae_only.json): no DiT LoRA. - [RGBA VAE + DiT LoRA](workflows/with_lora.json): default LoRA strength `1.0`. Describe the subject, appearance, pose, and style. You can append `transparent background`. Connect **VAE Decode → Save Image** to save the RGBA PNG. Full installation and node guide: [English](https://github.com/Chenkin-x/ComfyUI-CKMoss-Image-RGBA/blob/main/README.en.md) · [中文](https://github.com/Chenkin-x/ComfyUI-CKMoss-Image-RGBA). ## Showcase Actual outputs from the RGBA VAE and DiT LoRA, preserving the original alpha. Background colors and checkerboards are for display only. Actual CKMoss-Image-RGBA outputs

[Original transparent PNG files](samples) · [Prompts and seeds](samples/manifest.json) ## Credits and license Built on [MOSS-Image](https://github.com/OpenMOSS/MOSS-Image) and the [FLUX.2 VAE](https://github.com/black-forest-labs/flux2#flux2-autoencoder). Thanks to [niangao2331](https://github.com/niangao2331) for data support. This project uses [GPLv3](LICENSE). Upstream attribution: [NOTICE](NOTICE).