"""Z-Image-Turbo OpenVINO INT4 inference example (optimum-intel). Requires: diffusers==0.37.1, transformers==4.57.6, optimum-intel, openvino, nncf Export: optimum-cli export openvino -m Tongyi-MAI/Z-Image-Turbo --task text-to-image --weight-format fp16 ./fp16 Quantize: see quantize_int4.py (OVWeightQuantizationConfig bits=4, group 128, fallback adjust) """ import time import torch from optimum.intel import OVDiffusionPipeline MODEL_ID = "HelloSun/Z-Image-Turbo-OpenVINO-INT4" # or local path ./z-image-turbo-ov-int4 def main(): pipe = OVDiffusionPipeline.from_pretrained(MODEL_ID, compile=True) prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" # Turbo: num_inference_steps=9 => 8 DiT forwards (official README), guidance_scale=0.0 image = pipe( prompt=prompt, height=1024, width=1024, num_inference_steps=9, guidance_scale=0.0, generator=torch.Generator().manual_seed(42), ).images[0] image.save("z_image_turbo_ov_int4.png") print("saved z_image_turbo_ov_int4.png") if __name__ == "__main__": main()