Z-Image-Turbo_clear

• Guide (External site): English | Japanese
• Civitai
• Z-Image_clear_photoreal
• Z-Image_clear_anime
• Z-Image_clear_vae

Compare to Z-Image-Turbo (Original)

Z-Image-Turbo_clear | Z-Image-Turbo

The Z-Image-Turbo_clear model is a fine-tuned version of Z-Image-Turbo that improves details compared to the original.

For better results, please use it in conjunction with the dedicated VAE, Z-Image_clear_vae.

---

Model Format and Quality

Generally, model size and image quality are proportional.

• FP32 > FP16 ≒ BF16 > MXFP8 > FP8_scaled > NVFP4 > FP8

Generation time varies greatly depending on whether the model can be loaded into VRAM, and the supported formats are determined by the GPU generation.

	FP32	FP16	BF16	FP8	FP8_scaled	MXFP8	FP4
NVIDIA
RTX 5000 (Blackwell)	✅	✅	✅	✅	✅	✅	✅
RTX 4000 (Ada Lovelace)	✅	✅	✅	✅	✅	❌	❌
RTX 3000 (Ampere)	✅	✅	✅	❌	❌	❌	❌
RTX 2000 (Turing)	✅	✅	❌	❌	❌	❌	❌
GTX 1000 (Pascal)	✅	⚠️	❌	❌	❌	❌	❌
AMD
RX 9000 (RDNA4)	✅	✅	✅	✅	✅	❌	❌
RX 7000 (RDNA3)	✅	✅	✅	❌	❌	❌	❌
RX 6000 (RDNA2)	✅	✅	❌	❌	❌	❌	❌
RX 5000 (RDNA1)	✅	✅	❌	❌	❌	❌	❌
Intel
Arc B (Battlemage)	✅	✅	✅	❌	❌	❌	❌
Arc A (Alchemist)	✅	✅	✅	❌	❌	❌	❌

MXFP8 and NVFP4 are the latest standards designed for the RTX 5000 series. On unsupported GPUs, however, fallback mechanisms will actually slow down processing, so their use is not highly recommended.

For a comparison of image quality and generation time by format, please see this page.

---

Workflow

 

This PNG image contains a ComfyUI workflow.

---

Recomended Settings

• Sampler:
  • euler
  • res_multistep
• Scheduler: beta
• Steps: 9
• ModelSamplingAuraFlow: shift 7.5 (3-9)

---

Upscaler

your choice, examples:

• 4x-SwinIR-L_GAN
• 4x_NMKD-Superscale-SP_178000_G

---

Tips (External Sites)

• FP / BF Format Image Quality and Generation Time: English | Japanese
• ComfyUI VRAM management: English | Japanese
• Color Correct custom nodes in ComfyUI: English | Japanese
• About VAE: English | Japanese

---

License

• Apache-2.0