添加一个gradio界面，并修改快速开始

README.md

@@ -25,66 +25,25 @@ We are thrilled to release **Qwen-Image**, a 20B MMDiT image foundation model th
 - Due to heavy traffic, if you'd like to experience our demo online, we also recommend visiting DashScope, WaveSpeed, and LibLib. Please find the links below in the community support.
 
 ## Quick Start
-
-1. Make sure your transformers>=4.51.3 (Supporting Qwen2.5-VL)
-
-2. Install the latest version of diffusers
 ```
-pip install git+https://github.com/huggingface/diffusers
+git clone https://github.com/QwenLM/Qwen-Image
+cd Qwen-Image
+conda create -n Qwen-Image python=3.10
+conda activate Qwen-Image
+pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu128
+pip install -r requirements.txt
 ```
-
-The following contains a code snippet illustrating how to use the model to generate images based on text prompts:
-
-```python
-from diffusers import DiffusionPipeline
-import torch
-
-model_name = "Qwen/Qwen-Image"
-
-# Load the pipeline
-if torch.cuda.is_available():
-    torch_dtype = torch.bfloat16
-    device = "cuda"
-else:
-    torch_dtype = torch.float32
-    device = "cpu"
-
-pipe = DiffusionPipeline.from_pretrained(model_name, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-positive_magic = {
-    "en": "Ultra HD, 4K, cinematic composition.", # for english prompt
-    "zh": "超清，4K，电影级构图" # for chinese prompt
-}
-
-# Generate image
-prompt = '''A coffee shop entrance features a chalkboard sign reading "Qwen Coffee 😊 $2 per cup," with a neon light beside it displaying "通义千问". Next to it hangs a poster showing a beautiful Chinese woman, and beneath the poster is written "π≈3.1415926-53589793-23846264-33832795-02384197".'''
-
-negative_prompt = " " # Recommended if you don't use a negative prompt.
-
-
-# Generate with different aspect ratios
-aspect_ratios = {
-    "1:1": (1328, 1328),
-    "16:9": (1664, 928),
-    "9:16": (928, 1664),
-    "4:3": (1472, 1140),
-    "3:4": (1140, 1472)
-}
-
-width, height = aspect_ratios["16:9"]
-
-image = pipe(
-    prompt=prompt + positive_magic["en"],
-    negative_prompt=negative_prompt,
-    width=width,
-    height=height,
-    num_inference_steps=50,
-    true_cfg_scale=4.0,
-    generator=torch.Generator(device="cuda").manual_seed(42)
-).images[0]
-
-image.save("example.png")
+Download the model weights:
+```
+python download.py
+```
+Quick start with the gradio demo. Cost vram 39G:
+```
+python app.py
+```
+Quick start with the low vram mode. Cost vram 22G:
+```
+python app.py --vram low
 ```
 
 ## Show Cases

app.py

@@ -0,0 +1,123 @@
+import torch
+import argparse
+import os
+import numpy as np
+import datetime
+import random
+from diffusers import DiffusionPipeline
+import gradio as gr
+from optimum.quanto import freeze, qint8, quantize
+
+parser = argparse.ArgumentParser() 
+parser.add_argument("--server_name", type=str, default="127.0.0.1", help="IP地址，局域网访问改为0.0.0.0")
+parser.add_argument("--server_port", type=int, default=7891, help="使用端口")
+parser.add_argument("--share", action="store_true", help="是否启用gradio共享")
+parser.add_argument("--mcp_server", action="store_true", help="是否启用mcp服务")
+parser.add_argument('--vram', type=str, default='high', choices=['low', 'high'], help='显存模式')
+parser.add_argument('--lora', type=str, default="None", help='lora模型路径')
+args = parser.parse_args()
+
+
+if torch.cuda.is_available():
+    device = "cuda" 
+    if torch.cuda.get_device_capability()[0] >= 8:
+        dtype = torch.bfloat16
+    else:
+        dtype = torch.float16
+else:
+    device = "cpu"
+    dtype = torch.float32
+
+
+MAX_SEED = np.iinfo(np.int32).max
+os.makedirs("outputs", exist_ok=True)
+model_id = "models/Qwen-Image"
+pipe = DiffusionPipeline.from_pretrained(
+    model_id, 
+    torch_dtype=dtype,
+)
+if args.lora!="None":
+    pipe.load_lora_weights(args.lora)
+    print(f"加载{args.lora}")
+if args.vram=="high":
+    pipe.vae.enable_tiling()
+    pipe.enable_model_cpu_offload()
+else:
+    quantize(pipe.transformer, qint8)
+    freeze(pipe.transformer)
+    pipe.vae.enable_tiling()
+    pipe.enable_model_cpu_offload()
+
+
+def generate(
+    prompt,
+    negative_prompt,
+    width,
+    height,
+    num_inference_steps,
+    true_cfg_scale, 
+    seed_param,
+):
+    global pipe, model
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
+    if seed_param<0:
+        seed = random.randint(0, MAX_SEED)
+    else:
+        seed = seed_param
+    image = pipe(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        width=width,
+        height=height,
+        num_inference_steps=num_inference_steps,
+        true_cfg_scale=true_cfg_scale,
+        generator=torch.Generator().manual_seed(seed)
+    ).images[0]
+    image.save(f"outputs/{timestamp}.png")
+    return f"outputs/{timestamp}.png", seed
+
+
+with gr.Blocks(theme=gr.themes.Base()) as demo:
+    gr.Markdown("""
+            <div>
+                <h2 style="font-size: 30px;text-align: center;">Qwen-Image</h2>
+            </div>
+            """)
+    with gr.TabItem("Qwen-Image文生图"):
+        with gr.Row():
+            with gr.Column():
+                prompt = gr.Textbox(label="提示词", value="超清，4K，电影级构图，")
+                negative_prompt = gr.Textbox(label="负面提示词", value="")
+                width = gr.Slider(label="宽度（推荐1328x1328、1664x928、1472x1140）", minimum=256, maximum=2656, step=32, value=1328)
+                height = gr.Slider(label="高度", minimum=256, maximum=2656, step=32, value=1328)
+                num_inference_steps = gr.Slider(label="采样步数", minimum=1, maximum=100, step=1, value=50)
+                true_cfg_scale = gr.Slider(label="true cfg scale", minimum=1, maximum=10, step=0.1, value=4.0)
+                seed_param = gr.Number(label="种子，请输入正整数，-1为随机", value=-1)
+                generate_button = gr.Button("🎬 开始生成", variant='primary')
+            with gr.Column():
+                image_output = gr.Image(label="生成图片")
+                seed_output = gr.Textbox(label="种子")
+
+    gr.on(
+        triggers=[generate_button.click, prompt.submit, negative_prompt.submit],
+        fn = generate,
+        inputs = [
+            prompt,
+            negative_prompt,
+            width,
+            height,
+            num_inference_steps,
+            true_cfg_scale, 
+            seed_param,
+        ],
+        outputs = [image_output, seed_output]
+    )
+
+if __name__ == "__main__": 
+    demo.launch(
+        server_name=args.server_name, 
+        server_port=args.server_port,
+        share=args.share, 
+        mcp_server=args.mcp_server,
+        inbrowser=True,
+    )

download.py

@@ -0,0 +1,32 @@
+from huggingface_hub import snapshot_download, login
+import os
+
+model_name="Qwen/Qwen-Image"
+folder_name = model_name.split('/')[-1]
+
+def download_model(model_name):
+    snapshot_download(
+        repo_id=model_name,
+        local_dir=f"models/{folder_name}",
+        allow_patterns=[
+            "*",
+        ],
+        resume_download=True
+    )
+    print("下载成功！")
+
+try:
+    download_model(model_name)
+
+except Exception as e:
+    if "401" in str(e):
+        # 2. 需要登录
+        print("需要 Hugging Face 登录，创建新令牌（选择'Read'权限即可）")
+        token = input("请输入您的 Hugging Face 访问令牌 (https://huggingface.co/settings/tokens): ")
+
+        # 登录并重试
+        login(token=token)
+        download_model(model_name)
+
+    else:
+        print(f"下载失败: {str(e)}")

requirements.txt

@@ -0,0 +1,7 @@
+git+https://github.com/huggingface/diffusers
+transformers
+huggingface_hub
+gradio
+numpy==1.26.4
+accelerate
+optimum-quanto