Z-Image_clear_photoreal

• Guide (External Site): English | Japanese

• Hugging Face

• Z-Image-Turbo_clear

• Z-Image_clear_vae

Compare to Z-Image-Turbo_clear (distilled)

Z-Image_clear_photoreal | Z-Image-Turbo_clear (distilled)

Z-Image_clear_photoreal is a merged model of Z-Image-Turbo_clear and Z-Image (Base).

It maintains the high speed and stability of Z-Image-Turbo_clear while significantly improving image quality and diversity.

Image diversity

Z-Image_clear_photoreal

Z-Image_clear_photoreal produces rich variation in faces and compositions.

Z-Image-Turbo_clear (distilled)

Z-Image-Turbo_clear (distilled) tends to generate very similar-looking images with little diversity.

Model Formats and Image Quality

Generally, the image quality ranks in the following order:

FP32 > FP16 ≒ BF16 > INT8_ConvRot_HQ > MXFP8 > FP8_svd_scaled > FP8

• FP32 is mainly recommended for training purposes.

• FP16 / BF16 can generate high-quality illustrations.

• INT8_ConvRot_HQ offers an excellent balance of speed and quality while using roughly half the VRAM of FP16 / BF16.

• MXFP8, FP8_svd_scaled, and FP8 are currently not recommended, as they are inferior to INT8_ConvRot_HQ in both speed and image quality.

For a detailed comparison of each format, please refer to this page.

VAE

For the Z-Image_clear_photoreal model, we recommend using Z-Image_clear_vae.

• Z-Image_natural_vae: lower saturation, more natural look

• Z-Image_clear_vae: higher saturation, vivid and vibrant look

We provide several color-style variations, so feel free to choose the one that matches your preference.

Workflow

The PNG workflow can be downloaded from Hugging Face.

Recommended Settings

General

• Sampler: euler

• Scheduler: beta

• Steps: 12 (6-18)

• ModelSamplingAuraFlow: shift 6 (3-9)

• CFG scale: 1

Note: Negative Prompts have no effect.

Upscaler

your choice, examples:

• 4x-SwinIR-L_GAN

• 4x_NMKD-Superscale-SP_178000_G

Tips (External Sites)

• Z-Image-Turbo_clear: English | Japanese

• FP / BF Format Image Quality and Generation Time: English | Japanese

• ComfyUI Color Correction Nodes: English | Japanese

• FP precision: English | Japanese

License

• Apache-2.0