Descriptions:
This Veteran AI tutorial introduces a hybrid ComfyUI workflow that combines two members of the Z-Image model family — Z-Image Turbo and Z-Image Base — to achieve portrait image quality that neither model delivers alone. The core insight, sourced from a community post on Twitter, is that Turbo excels at fast composition and pleasing character aesthetics but loses fine skin detail, while Base produces rich detail but can appear overly clinical. By splitting the sampling pipeline between the two, the workflow captures the strengths of both.
The technical implementation uses an Advanced Sampler with a shared 20-step budget. The handover point is set to step 5 via the “End At Step” parameter: steps 0–5 run under Z-Image Turbo with CFG 1.0, and steps 5–20 hand off to Z-Image Base with CFG 4.0. A critical detail is that “Add Noise” must be enabled for both samplers — disabling it in the second pass prevents fine detail from forming. The shared seed between models ensures consistent noise distribution since both are from the same model family. Resolution is set to 1024×1536.
The presenter tested approximately 20 images across five quality dimensions — Oriental aesthetics, composition and atmosphere, light and shadow, skin texture and realism — and reports near-100% keeper rate without regeneration. The workflow is available on RunningHub for direct access and testing.
📺 Source: Veteran AI · Published February 03, 2026
🏷️ Format: Tutorial Demo
