ComfyUI nodes: a beginner's guide to connections

This beginner lesson explains how ComfyUI nodes form an image generation workflow. Each node handles a task, such as loading an image, encoding a prompt or selecting a model. The tutorial covers adding nodes through the right-click menu or searching by name after double-clicking the canvas.

Examples include Load Image, CLIP Text Encode and Load Diffusion Model. Reroute nodes organize connection lines without processing their data. The speaker also distinguishes built-in nodes from custom nodes downloaded from the community.

The connection guide explains that inputs sit on the left and outputs on the right. Compatible connectors pass data between nodes, while their colors identify data types. The lesson covers text, pixel images, models and latent data, then explains how VAE encoding and decoding connect images with latent representations. CLIP and conditioning receive separate explanations: one interprets prompts, while the other guides generation.

The speaker describes quantized GGUF models as useful when memory is limited and recommends safetensors over older .ckpt files because of potential malicious code. The final example introduces ControlNet for guiding output with reference poses, edges or depth maps. This is a guide to reading workflows, rather than an installation or hardware setup tutorial.