MOSS-Image Vision Text Encode (ComfyUI node + workflow)
retained text
Built a custom node for MOSS-Image that feeds reference images directly through the Qwen3.5-9B vision tower and leaves its embeddings in the text conditioning - so the image conditions generation as part of the prompt, not as a VAE latent. \- Custom Node + Workflow: [https://github.com/ThetaCursed/ComfyUI-MOSSImageTextEncoder](https://github.com/ThetaCursed/ComfyUI-MOSSImageTextEncoder) \- LoRA used: [https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/tree/main](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/tree/main)
document fetched 2026-10-22 00:34 · canonical https://reddit.com/r/StableDiffusion/comments/1xbjxcr/mossimage_vision_text_encode_comfyui_node_workflow