ToddEverett
commited on
Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
|
|
|
|
|
|
|
1 |
+
在 `gr.load()` 基础上,你无法直接通过简单的参数设置来更改图像尺寸。因此,这种情况需要将 `gr.load()` 的加载方式更改为 `from_pretrained`,这样你就可以通过 `width` 和 `height` 参数来调整生成图像的尺寸。
|
2 |
+
|
3 |
+
这里是修改后的代码示例,将图像的默认尺寸设置为 `1024x768`,并允许用户自定义长宽比。
|
4 |
+
|
5 |
+
```python
|
6 |
import gradio as gr
|
7 |
+
import torch
|
8 |
+
from diffusers import DiffusionPipeline
|
9 |
+
|
10 |
+
# 设置模型和设备
|
11 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
12 |
+
model_repo_id = "stabilityai/stable-diffusion-3.5-large"
|
13 |
+
|
14 |
+
# 加载模型
|
15 |
+
pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch.float16 if device == "cuda" else torch.float32)
|
16 |
+
pipe = pipe.to(device)
|
17 |
+
|
18 |
+
# 推理函数,允许自定义宽度和高度
|
19 |
+
def generate_image(prompt, width=1024, height=768, guidance_scale=7.5, num_inference_steps=50):
|
20 |
+
image = pipe(
|
21 |
+
prompt=prompt,
|
22 |
+
width=width,
|
23 |
+
height=height,
|
24 |
+
guidance_scale=guidance_scale,
|
25 |
+
num_inference_steps=num_inference_steps
|
26 |
+
).images[0]
|
27 |
+
return image
|
28 |
+
|
29 |
+
# 设置 Gradio 界面
|
30 |
+
with gr.Blocks() as demo:
|
31 |
+
gr.Markdown(" # Stable Diffusion 3.5 - Custom Width & Height")
|
32 |
+
|
33 |
+
with gr.Row():
|
34 |
+
prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt here")
|
35 |
+
|
36 |
+
# 允许用户设置宽度和高度
|
37 |
+
width = gr.Slider(label="Width", minimum=512, maximum=1024, step=64, value=1024)
|
38 |
+
height = gr.Slider(label="Height", minimum=512, maximum=1024, step=64, value=768)
|
39 |
+
|
40 |
+
guidance_scale = gr.Slider(label="Guidance Scale", minimum=0.1, maximum=15.0, step=0.1, value=7.5)
|
41 |
+
num_inference_steps = gr.Slider(label="Number of Inference Steps", minimum=1, maximum=100, step=1, value=50)
|
42 |
+
|
43 |
+
# 显示生成的图像
|
44 |
+
output_image = gr.Image(label="Generated Image")
|
45 |
+
|
46 |
+
# 点击按钮运行生成
|
47 |
+
generate_btn = gr.Button("Generate")
|
48 |
+
generate_btn.click(generate_image, inputs=[prompt, width, height, guidance_scale, num_inference_steps], outputs=output_image)
|
49 |
+
|
50 |
+
# 启动 Gradio 界面
|
51 |
+
demo.launch()
|
52 |
+
```
|
53 |
|
54 |
+
### 代码说明
|
55 |
+
- **模型加载**:将 `gr.load` 替换为 `DiffusionPipeline.from_pretrained`,这样可以手动控制图像尺寸。
|
56 |
+
- **`generate_image` 函数**:此函数接收 `width` 和 `height` 参数,允许用户生成不同长宽比的图像。
|
57 |
+
- **Gradio 界面**:提供 `Slider` 供用户自定义 `width` 和 `height`,并设置默认值为 `1024x768`。
|