Z-Image SD variety prefiner

Z-Image SD variety prefiner is for trying to generate more varied Z-Image-Turbo images: it starts encoding with SD 1.5, uses Z-Image-Turbo for most of the work and the final stage, takes a seed you can vary, and outputs an image.

SD 1.5 starts the encoding, while Z-Image-Turbo does most of the work and handles the final stage.

The description says the prefiner adds variety across the image, with the undescribed background seeming most affected.

The SD 1.5 stage generates a 512×512 image, and the author notes that it can be resized to most shapes for final generation.

A SaveImage step is listed in the workflow.

Published by Ken G6 on Civitai at https://civitai.com/models/2172045.

The listed model files are ae.safetensors, hddclip/qwen_3_4b.safetensors, sd15/realisticVisionV51_v51VAE.safetensors, and z_image_turbo_bf16.safetensors.

Listed node types include CheckpointLoaderSimple, CLIPLoader, CLIPTextEncode, EmptySD3LatentImage, KSamplerAdvanced, LatentUpscale, and ModelSamplingAuraFlow.

Other listed node types are PrimitiveInt, PrimitiveStringMultiline, SaveImage, UNETLoader, VAEDecode, VAEEncode, and VAELoader.

The author lists sgm-uniform as a preferred Z-Image scheduler instead of the workflow default.

Increasing SD 1.5 CFG can reduce patterns across the image, but not eliminate them.

A relatively high step break of 2 or 3 can sometimes produce a cartoony look, but the described effect is not reliably repeatable.

No license is listed for this workflow; review the Civitai page before reuse.

Suggestion · not verified

Before using it, confirm that your setup has the listed model files and node types.

Suggestion · not verified