example.py

"""VBVR-Pro-Qwen-Image-Edit inference example.

Usage:
    python example.py --model_path ./VBVR-Pro-Qwen-Image-Edit
"""

import argparse
import torch
from PIL import Image
from diffusers import QwenImageEditPlusPipeline

parser = argparse.ArgumentParser()
parser.add_argument("--model_path", type=str, default="VBVR-Pro-Qwen-Image-Edit")
parser.add_argument("--image", type=str, required=True, help="Path to input image")
parser.add_argument("--prompt", type=str, required=True, help="Editing instruction")
parser.add_argument("--output", type=str, default="output.png")
parser.add_argument("--steps", type=int, default=40)
parser.add_argument("--seed", type=int, default=42)
args = parser.parse_args()

print(f"Loading model from: {args.model_path}")
pipe = QwenImageEditPlusPipeline.from_pretrained(
    args.model_path, torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()

image = Image.open(args.image).convert("RGB")
print(f"Input image: {args.image} ({image.size[0]}x{image.size[1]})")

output = pipe(
    image=[image],
    prompt=args.prompt,
    num_inference_steps=args.steps,
    true_cfg_scale=4.0,
    negative_prompt=" ",
    guidance_scale=1.0,
    generator=torch.manual_seed(args.seed),
).images[0]

output.save(args.output)
print(f"Saved to: {args.output}")