DEV Community

AI-KSK
AI-KSK

Posted on

Z-Image Base plus Turbo: two-step generation and segmented rendering

Z-Image has two personalities. The Base model renders slowly and truthfully; the Turbo model renders fast and approximately. This episode is about using both in one workflow - and about segmented rendering, where the two are mixed across stages.

Watch the episode: https://youtu.be/Y6L5qkA8ZYs

The two workflows

  1. Two-step generation - the Turbo path pushed to its limit. Two sampling steps, usable results, near-instant turnaround.
  2. Base plus Turbo segmented render - the composition and the detail are produced in separate stages, with each stage running the model that suits it.

Why segmented rendering is interesting

Most acceleration is applied uniformly: the whole image gets the same cheap treatment. Segmented rendering instead says the layout can come from the fast model and the finish from the slow one. That is how you get a Turbo-speed sketch with Base-level texture.

What to watch for

  • Two steps is a sketch, not a finished image. Judge it at thumbnail size, not crop level.
  • Segmented renders are sensitive to where the segment boundary falls; a hard boundary shows.
  • The fast stage sets the composition, so a bad fast stage cannot be rescued by the slow one.

Honest limits

  • These are ComfyUI workflows with no bundled weights.
  • Very low step counts drift on faces first - if the subject is a person, check eyes before anything else.
  • Segmented rendering costs more total time than pure Turbo, less than pure Base.

Top comments (0)