ComfyUI tutorial: node graphs, Cloud and Desktop

This beginner tutorial explains ComfyUI through a text-to-image workflow in Comfy Cloud. It follows the connections between checkpoint loading, positive and negative text prompts, an empty latent image, KSampler, VAE decoding and image output. The presenter also covers navigation, node search and running a single output branch instead of the whole graph.

For readers interested in local AI, the deployment comparison matters: the speaker describes Comfy Desktop as an app that uses your own GPU and supports local storage and custom nodes. The walkthrough stays in Cloud, where users avoid installation but cannot access every community node or experimental model. It is not a Desktop installation guide.

Much of the lesson concerns keeping graphs readable. Reroutes reshape wires, Get/Set nodes pass named values across the graph, and frames group related work. Subgraphs package nodes into a smaller interface; App Mode exposes selected inputs and outputs for collaborators.

The presenter stresses matching CLIP, VAE and LoRA components to their base model, using Stable Diffusion 1.5 as an example. The tutorial distinguishes templates, blueprints and workflow JSON, then demonstrates export options. It also explains workflow metadata embedded in PNG images and notes that partner API nodes consume credits. ComfyUI is presented as open source, while the demonstrated Cloud environment has a curated selection of extensions.