Local Image Generation with ComfyUI

ComfyUI is a node-graph interface for running diffusion models locally — you wire up the pipeline (load model → encode prompt → sample → decode) as visual nodes, giving precise control over every step. theLAB uses it for local image and video generation (e.g. Wan-based image-to-video).

#Why node-based

Unlike one-box tools, ComfyUI exposes the actual pipeline, so you can:

• Swap samplers, schedulers, and VAEs independently.
• Chain ControlNet, LoRA, upscalers, and inpainting in one graph.
• Save a workflow as JSON and reproduce it exactly.

#Running it

git clone https://github.com/comfyanonymous/ComfyUI
pip install -r requirements.txt
python main.py --listen 0.0.0.0 --port 8189

#VRAM reality

Local image/video gen is slow but free — great for batch jobs that would be costly on hosted APIs. Offload to a GPU box on the fleet and queue work.

Related: ADI Montage — Agentic Video Studio · Serving Multiple Models on a Dual-GPU Box