ComfyUI node basics: build a Flux 2 Klein workflow

This beginner tutorial explains how ComfyUI nodes pass data through a workflow, then builds a Flux 2 Klein 9B text-to-image setup from an empty canvas. It covers inputs, outputs, matching wire colors and four ways to add nodes. Viewers also learn to edit settings, control the seed, arrange nodes and save a workflow for reuse.

The presenter uses the classic node appearance and says some custom nodes do not yet work well with Nodes 2.0. The lesson distinguishes bypassing a node from muting it and explains how missing custom node packs can produce red nodes. A newer core node may instead require a ComfyUI update.

The example connects a diffusion model, text encoder and VAE to prompt encoding, KSampler, decoding and image saving. It uses a 1024 × 1024 latent image, four sampling steps and CFG set to one. The presenter routes the prompt through Conditioning Zero Out for the negative input, explaining that Klein does not need a negative prompt.

The supplied description links three required model files and their destination folders, with an FP8 or INT8 choice for the diffusion model. The final demonstration changes the prompt, reruns the workflow and saves it with Ctrl+S; the presenter attributes the faster rerun to ComfyUI executing only changed nodes.