File size: 4,440 Bytes
a1eec31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fb08f6a
a1eec31
fb08f6a
a1eec31
 
 
 
 
fb08f6a
a1eec31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187b850
 
 
 
a1eec31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fb08f6a
a1eec31
 
 
8bf8763
a1eec31
187b850
8bf8763
a1eec31
 
8bf8763
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a1eec31
 
 
8bf8763
a1eec31
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
187b850
8bf8763
187b850
a1eec31
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
import torch
import gradio as gr
import torch
import os
from PIL import Image
from torch import autocast
from perpneg_diffusion.perpneg_stable_diffusion.pipeline_perpneg_stable_diffusion import PerpStableDiffusionPipeline

has_cuda = torch.cuda.is_available()
device = torch.device('cpu' if not has_cuda else 'cuda')
print(device)

# initialize stable diffusion model
pipe = PerpStableDiffusionPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    # use_auth_token=True
).to(device)

def dummy(images, **kwargs):
    return images, False


pipe.safety_checker = dummy

examples = [
    [
        "an armchair in the shape of an avocado | cushion in the armchair",
        "1 | -0.3",
        "145",
        "7.5"
    ],
    [
        "an armchair in the shape of an avocado",
        "1",
        "145",
        "7.5"
    ],
    [
        "a peacock, back view | a peacock, front view",
        "1 | -3.5",
        "30",
        "7.5"
    ],
    [
        "a peacock, back view",
        "1",
        "30",
        "7.5"
    ],    
    [
        "A boy wearing sunglasses | a pair of sunglasses with white frame",
        "1 | -0.35",
        "200",
        "11"
    ],
    [
        "A boy wearing sunglasses",
        "1",
        "200",
        "11",
    ],
    [
        "a photo of an astronaut riding a horse | a jumping horse | a white horse", 
        "1 | -0.3 | -0.1",
        "1988",
        "10"
    ],             
    [
        "a photo of an astronaut riding a horse | a jumping horse", 
        "1 | -0.3",
        "1988",
        "10"
    ],
    [
        "a photo of an astronaut riding a horse", 
        "1",
        "1988",
        "10"
    ],         
]







def predict(prompt, weights, seed, scale=7.5, steps=50):
    try:
        with torch.no_grad():
            has_cuda = torch.cuda.is_available()
            with autocast('cpu' if not has_cuda else 'cuda'):
                if has_cuda:
                    generator = torch.Generator('cuda').manual_seed(int(seed))
                else:
                    generator = torch.Generator().manual_seed(int(seed))
                image_perpneg = pipe(prompt, guidance_scale=float(scale), generator=generator,
                            num_inference_steps=steps, weights=weights)["images"][0]
                return image_perpneg
    except Exception as e:
        print(e)
        return None






app = gr.Blocks()
with app:
    # gr.Markdown(
    #     "# **<p align='center'>AMLDS Video Tagging</p>**"
    # )
    gr.Markdown(
        "# **<p align='center'>Perp-Neg: Iterative Editing and Robust View Generation Using Stable Diffusion</p>**"
    )    
    gr.Markdown(
        """
        ### **<p align='center'>Demo created by Huangjie Zheng and Reza Armandpour</p>**
        """
    )
 
    with gr.Row():
        with gr.Column():
            # with gr.Tab(label="Inputs"):
            # gr.Markdown(
            #     "### Prompts (a list of prompts separated by vertical bar | )"
            # )
            prompt = gr.Textbox(label="Prompts (a list of prompts separated by vertical bar | ):", show_label=True, placeholder="a peacock, back view | a peacock, front view")
            weights = gr.Textbox(
                label="Weights (a list of weights separated by vertical bar | )", show_label=True, placeholder="1 | -3.5"
            )
            seed = gr.Textbox(
                label="Seed", show_label=True, value=30
            )
            scale = gr.Textbox(
                label="Guidance scale", show_label=True, value=7.5
            )                   
            image_gen_btn = gr.Button(value="Generate")

        with gr.Column():
            img_output = gr.Image(
                label="Result",
                show_label=True,
            )


    gr.Markdown("**Examples:**")
    gr.Examples(
        examples,
        [prompt, weights, seed, scale],
        [img_output],
        fn=predict,
        cache_examples=False,
    )

    image_gen_btn.click(
        predict,
        inputs=[prompt, weights, seed, scale],
        outputs=[img_output],
    )
    gr.Markdown("""
    \n The algorithem is based on the paper: [Re-imagine the Negative Prompt Algorithm: Transform 2D Diffusion into 3D, alleviate Janus problem and Beyond.](https://Perp-Neg.github.io).
    """)    

    gr.Markdown(
        """
        \n Demo created by: Huangjie Zheng and Reza Armandpour</a>.
        """
    )

app.launch()