Z-Image：阿里开源高效图片生成模型

# 1. 环境准备
python -m venv zimage-env
source zimage-env/bin/activate  # Windows用：zimage-env\Scripts\activate

# 2. 安装依赖
pip install --upgrade "diffusers[torch]" transformers accelerate safetensors

# 3. 安装GPU版PyTorch（根据CUDA版本在pytorch.org复制命令）

# 4. Python代码示例
from diffusers import DiffusionPipeline
import torch

# 加载Z-Image Turbo模型
model_id = "Tongyi-MAI/Z-Image-Turbo"  # 官方模型ID
pipe = DiffusionPipeline.from_pretrained(
    model_id,
    torch_dtype=torch.float16
).to("cuda")

# 生成图像
prompt = "a cinematic concept art of a futuristic city at sunset, ultra detailed, 4k"
image = pipe(
    prompt,
    num_inference_steps=25,   # 步数，20-30一般够用
    guidance_scale=7.5        # 类似CFG Scale，5-9之间试
).images[0]

image.save("zimage_demo.png")

# 5. 性能优化（可选）
pipe.enable_xformers_memory_efficient_attention()  # 节省显存
torch.backends.cudnn.benchmark = True  # 同一尺寸多次生成时加速

# 6. Flask风格API示例 - 完整可用
from flask import Flask, request, send_file
from io import BytesIO

app = Flask(__name__)

# 初始化模型（只加载一次）
model_id = "Tongyi-MAI/Z-Image-Turbo"
pipe = DiffusionPipeline.from_pretrained(
    model_id,
    torch_dtype=torch.float16
).to("cuda")

@app.route("/generate", methods=["POST"])
def generate():
    data = request.json
    prompt = data.get("prompt", "a beautiful painting")
    steps = int(data.get("steps", 25))
    guidance = float(data.get("guidance", 7.5))

    image = pipe(
        prompt,
        num_inference_steps=steps,
        guidance_scale=guidance
    ).images[0]

    buf = BytesIO()
    image.save(buf, format="PNG")
    buf.seek(0)
    return send_file(buf, mimetype="image/png")

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=8000)