ComfyUI lab 9: Control Z-Image with line structure

Line control divides an image problem into two channels: a structural signal says where boundaries belong, while text says what the scene should become. That separation is why sketches can survive large changes in material and lighting.

Two strengths, two jobs

Control conditioning scale governs how strongly the generated image follows extracted lines. Text guidance governs pressure toward the prompt. They are not interchangeable. Sweep control strength at a fixed seed—0.45, 0.75, 1.05—before touching guidance. Low values permit invention; high values can trace noise or flatten natural detail.

The open node prefers the CuteDSL Z-Image path when installed and can fall back to compatible Diffusers components. It requires a declared ZIMAGE_CONTROLNET_DIR; the workflow does not pretend that experimental weights appear automatically. The source image is resized to valid model dimensions and converted to line art with an explicit fallback extractor.

Prepare useful structure

Clean sketches beat photographs full of texture. Remove text, crop to the intended composition, and ensure the important silhouette has contrast. If a line is not meaningful to the design, remove it before asking the model to obey it.

Latent teleportation remains research, not a magic speed checkbox. Measure the shipped CuteDSL or Diffusers path with the exact model, steps, resolution, and GPU you use.

Run the lab

• Open this graph in Comfy Studio
• Download the exact API-format workflow
• CLI: app comfy download zimage-line-control

Record the graph JSON, input assets, seed where relevant, model or service version, custom-node commit, and final artifact together. That bundle—not a screenshot—is the reproducible creative unit.

Open-source map

• Open CuteDSL nodes — MIT accelerated Z-Image integration
• ComfyUI ControlNet examples — Core structural-conditioning examples

Creative Systems Labs: 9 of 12. Previous: Prototype frame transforms safely Next: Restyle without losing structure