ComfyUI-DreamLite

Experimental ComfyUI custom node wrapper for ByteVisionLab/DreamLite.

This is not an official ByteDance/ByteVisionLab project. It wraps the official DreamLite Python pipelines instead of integrating DreamLite into ComfyUI's native KSampler stack.

Nodes

• DreamLite Generate/Edit: text-to-image and text-guided image editing through the official pipeline API.

Install

1. Copy this folder to ComfyUI/custom_nodes/ComfyUI-DreamLite.

2. Install dependencies into the same Python environment used by ComfyUI:

   pip install -r ComfyUI/custom_nodes/ComfyUI-DreamLite/requirements.txt
   

3. Clone the official DreamLite repository somewhere locally:

   git clone https://github.com/ByteVisionLab/DreamLite.git
   

4. Put DreamLite model weights in the official repo layout, for example:

   DreamLite/models/DreamLite-base
   DreamLite/models/DreamLite-mobile
   

5. Restart ComfyUI.

6. Add DreamLite > DreamLite Generate/Edit.

7. Set dreamlite_repo to your local DreamLite repository path.

8. Set model_path to models/DreamLite-base, models/DreamLite-mobile, or an absolute model folder path.

Recommended settings

• Base: variant=base, steps=0 or steps=28, model_path=models/DreamLite-base.
• Mobile: variant=mobile, steps=0 or steps=4, model_path=models/DreamLite-mobile.

Notes

• Optional image input enables text-guided editing.
• size_mode=use_input_image_size uses the connected image's dimensions for editing.
• steps=0 uses official defaults: base=28 and mobile=4.
• This node caches the pipeline in memory. Restart ComfyUI to fully unload it.
• The wrapper has not been validated without access to the gated DreamLite weights.