ComfyUI from zero: the seven nodes of a basic workflow

ComfyUI is a node-based interface for diffusion models, at https://github.com/comfyanonymous/ComfyUI It looks intimidating, but the default workflow is seven nodes, and every advanced workflow is those seven plus additions. The seven nodes 1. Load Checkpoint: loads the model, and outputs MODEL, CLIP and VAE. 2. CLIP Text Encode (positive): your prompt. 3. CLIP Text Encode (negative): your negative prompt. 4. Empty Latent Image: sets width, height and batch size. 5. KSampler: seed, steps, CFG, sampler, scheduler, denoise. This is where the image is made. 6. VAE Decode: turns the latent into pixels. 7. Save Image: writes the PNG, with the whole workflow embedded in it. Install The project README covers Windows (a portable build), Linux and macOS. Models go in ComfyUI/models/checkpoints, LoRAs in models/loras, ControlNets in models/controlnet. Two habits that save hours - Drag any ComfyUI-made PNG onto the canvas and the workflow that made it loads. Share PNGs, not screenshots. - Install ComfyUI-Manager to install missing custom nodes when you load someone else's workflow. Extending the seven - img2img: replace Empty Latent with Load Image, then VAE Encode, and set KSampler denoise below 1. - LoRA: put a Load LoRA node between Load Checkpoint and everything that uses MODEL and CLIP. - ControlNet: Apply ControlNet between your positive CLIP Text Encode and the KSampler. The wiki's overview: https://wiki.soapbox.community/wiki/Generative_Image_AI_and_ComfyUI