Skip to content

README.md

Latest commit

 

History

History
35 lines (22 loc) · 1.51 KB

README.md

File metadata and controls

35 lines (22 loc) · 1.51 KB
Raw

Z-Image-Turbo

English | 简体中文

Prepare

Complete the environment setup and download the complete Z-Image-Turbo model directory.

Task Shell entry
Text to image z_image_turbo_t2i.sh
Text to image, two GPUs z_image_turbo_t2i_dist.sh
Image to image z_image_turbo_i2i.sh

Set lightx2v_path, model_path, and CUDA_VISIBLE_DEVICES in the selected script. Edit the prompt and, for image-to-image generation, image_path and --i2i_denoise_strength.

Run

Run from the repository root:

bash scripts/z_image/z_image_turbo_t2i.sh

The basic text-to-image example saves to save_results/z_image_turbo.png; change --save_result_path to choose another location. Sampling settings are in the JSON selected by --config_json under configs/z_image.

Python API

z_image_turbo.py uses FP8 DiT weights and CPU offload. Also download z_image_turbo_scaled_fp8_e4m3fn.safetensors from Z-Image-Turbo-Quantized, then edit model_path, dit_quantized_ckpt, and save_result_path in the example.

python scripts/z_image/inference/z_image_turbo.py

See the Python API guide for shared usage.