ComfyUI: build a Z-Image Turbo text-to-image workflow

This tutorial builds a ComfyUI text-to-image workflow from a blank canvas and explains the connections between its nodes. It assumes ComfyUI already runs locally or on RunPod and that the required model files are installed. It does not cover environment installation.

The walkthrough separates the diffusion model, VAE and text encoder loaders, then connects prompt conditioning and an empty latent image to KSampler. VAE Decode converts the sampler's latent output into pixels, and Save Image receives the result. This gives viewers a practical introduction to local AI image generation through a node graph.

For the Z-Image Turbo example, the speaker adds Conditioning Zero Out to supply KSampler's negative conditioning input, explaining that this model does not accept negative prompts. The demonstrated settings use CFG 1 and eight sampling steps, with an initial image width of 512 pixels and height of 1024 pixels.

Later examples use a higher resolution. The speaker warns that larger images can exhaust GPU memory and recommends smaller dimensions when VRAM or RAM is limited. Claude also appears as a prompt-writing aid, and the speaker recommends it for download commands and troubleshooting. Those recommendations are advice from the presenter, rather than verified guarantees of accurate fixes.