Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import random
|
2 |
+
import spaces
|
3 |
+
import torch
|
4 |
+
import gradio as gr
|
5 |
+
|
6 |
+
from modeling.dmm_pipeline import StableDiffusionDMMPipeline
|
7 |
+
from huggingface_hub import snapshot_download
|
8 |
+
|
9 |
+
|
10 |
+
ckpt_path = "ckpt"
|
11 |
+
snapshot_download(repo_id="MCG-NJU/DMM", local_dir=ckpt_path)
|
12 |
+
|
13 |
+
pipe = StableDiffusionDMMPipeline.from_pretrained(
|
14 |
+
ckpt_path,
|
15 |
+
torch_dtype=torch.float16,
|
16 |
+
use_safetensors=True
|
17 |
+
)
|
18 |
+
pipe.to("cuda")
|
19 |
+
|
20 |
+
|
21 |
+
@spaces.GPU
|
22 |
+
def generate(prompt: str,
|
23 |
+
negative_prompt: str,
|
24 |
+
model_id: int,
|
25 |
+
seed: int = 1234,
|
26 |
+
height: int = 512,
|
27 |
+
width: int = 512,
|
28 |
+
all: bool = True):
|
29 |
+
if all:
|
30 |
+
outputs = []
|
31 |
+
for i in range(pipe.unet.get_num_models()):
|
32 |
+
output = pipe(
|
33 |
+
prompt=prompt,
|
34 |
+
negative_prompt=negative_prompt,
|
35 |
+
width=width,
|
36 |
+
height=height,
|
37 |
+
num_inference_steps=25,
|
38 |
+
guidance_scale=7,
|
39 |
+
model_id=i,
|
40 |
+
generator=torch.Generator().manual_seed(seed),
|
41 |
+
).images[0]
|
42 |
+
outputs.append(output)
|
43 |
+
return outputs
|
44 |
+
else:
|
45 |
+
output = pipe(
|
46 |
+
prompt=prompt,
|
47 |
+
negative_prompt=negative_prompt,
|
48 |
+
width=width,
|
49 |
+
height=height,
|
50 |
+
num_inference_steps=25,
|
51 |
+
guidance_scale=7,
|
52 |
+
model_id=int(model_id),
|
53 |
+
generator=torch.Generator().manual_seed(seed),
|
54 |
+
).images[0]
|
55 |
+
return [output,]
|
56 |
+
|
57 |
+
|
58 |
+
candidates = [
|
59 |
+
"0. [JuggernautReborn] realistic",
|
60 |
+
"1. [MajicmixRealisticV7] realistic, Asia portrait",
|
61 |
+
"2. [EpicRealismV5] realistic",
|
62 |
+
"3. [RealisticVisionV5] realistic",
|
63 |
+
"4. [MajicmixFantasyV3] animation",
|
64 |
+
"5. [MinimalismV2] illustration",
|
65 |
+
"6. [RealCartoon3dV17] cartoon 3d",
|
66 |
+
"7. [AWPaintingV1.4] animation",
|
67 |
+
]
|
68 |
+
|
69 |
+
def main():
|
70 |
+
with gr.Blocks() as demo:
|
71 |
+
gr.Markdown(
|
72 |
+
"""
|
73 |
+
# DMM Demo
|
74 |
+
The checkpoint is https://huggingface.co/MCG-NJU/DMM.
|
75 |
+
"""
|
76 |
+
)
|
77 |
+
with gr.Row():
|
78 |
+
with gr.Column():
|
79 |
+
with gr.Column():
|
80 |
+
model_id = gr.Dropdown(candidates, label="Model Index", type="index")
|
81 |
+
all_check = gr.Checkbox(label="All (ignore the selection above)")
|
82 |
+
prompt = gr.Textbox("portrait photo of a girl, long golden hair, flowers, best quality", label="Prompt")
|
83 |
+
negative_prompt = gr.Textbox("worst quality,low quality,normal quality,lowres,watermark,nsfw", label="Negative Prompt")
|
84 |
+
with gr.Row():
|
85 |
+
seed = gr.Number(0, label="Seed", precision=0, scale=3)
|
86 |
+
update_seed_btn = gr.Button("🎲", scale=1)
|
87 |
+
with gr.Row():
|
88 |
+
height = gr.Number(768, step=8, label="Height (suggest 512~768)")
|
89 |
+
width = gr.Number(512, step=8, label="Width")
|
90 |
+
submit_btn = gr.Button("Submit", variant="primary")
|
91 |
+
output = gr.Gallery(label="images")
|
92 |
+
|
93 |
+
submit_btn.click(generate,
|
94 |
+
inputs=[prompt, negative_prompt, model_id, seed, height, width, all_check],
|
95 |
+
outputs=[output])
|
96 |
+
update_seed_btn.click(lambda: random.randint(0, 1000000),
|
97 |
+
outputs=[seed])
|
98 |
+
|
99 |
+
demo.launch()
|
100 |
+
|
101 |
+
|
102 |
+
if __name__ == "__main__":
|
103 |
+
main()
|