File size: 4,390 Bytes
4a7978f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7239d24
4a7978f
 
 
 
b7cfa06
7239d24
2ebe433
4a7978f
 
5da027f
 
 
b7cfa06
7239d24
2ebe433
7239d24
 
 
 
 
 
 
2ebe433
5da027f
 
 
7239d24
5da027f
b7cfa06
7239d24
2ebe433
5da027f
 
 
7239d24
5da027f
b7cfa06
7239d24
2ebe433
5da027f
 
4a7978f
 
 
 
 
 
7540b2d
4a7978f
 
 
 
 
 
 
 
 
 
 
 
d02d8ed
4a7978f
7540b2d
7239d24
4a7978f
 
dc6ccc0
4a7978f
 
7239d24
48c650b
7239d24
 
 
 
4a7978f
 
 
 
 
 
 
 
 
 
 
 
 
a7a89c0
4a7978f
 
 
 
 
 
 
 
 
c89ea6c
4a7978f
 
 
 
37d3c52
 
 
 
 
 
 
4a7978f
 
 
 
 
 
 
 
 
 
 
 
7239d24
5da027f
4a7978f
 
 
 
 
 
 
 
 
 
 
 
 
 
7239d24
4a7978f
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import math
import random

import gradio as gr
import torch
from PIL import Image, ImageOps
from diffusers import StableDiffusionSAGPipeline


help_text = """
"""



examples = [
    [
        ' ',
        50,
        "Fix Seed",
        35934,
        3.0,
        1.0,
    ],
    [
        '.',
        50,
        "Fix Seed",
        24865,
        3.0,
        1.0,
    ],
    [
        'A poster',
        50,
        "Fix Seed",
        37956,
        3.0,
        1.0,
    ],
    [
        'A high-quality living room',
        50,
        "Fix Seed",
        78710,
        3.0,
        1.0,
    ],
    [
        'A Scottish Fold playing with a ball',
        50,
        "Fix Seed",
        11511,
        3.0,
        1.0,
    ],
]


model_id = "runwayml/stable-diffusion-v1-5"

def main():
    pipe = StableDiffusionSAGPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to('cuda')

    def generate(
        prompt: str,
        steps: int,
        randomize_seed: bool,
        seed: int,
        cfg_scale: float,
        sag_scale: float,
    ):
        seed = random.randint(0, 100000) if randomize_seed else seed

        generator = torch.manual_seed(seed)
        ori_image = pipe(prompt, generator=generator, guidance_scale=cfg_scale, sag_scale=0.0).images[0]
        generator = torch.manual_seed(seed)
        sag_image = pipe(prompt, generator=generator, guidance_scale=cfg_scale, sag_scale=sag_scale).images[0]
        return [ori_image, sag_image, seed]

    def reset():
        return [0, "Randomize Seed", 42, 3.0, 0.75, None, None]

    with gr.Blocks() as demo:
        gr.HTML("""<h1 style="font-weight: 900; margin-bottom: 10px;">
            Self-Attention Guidance Demo
            </h1>
            <h3 style="font-weight: 500; margin-bottom: 10px;">
            Condition-Agnostic Diffusion Guidance Using the Internal Self-Attention
            </h3>
        """)
        with gr.Row():
            with gr.Column(scale=5):
                prompt = gr.Textbox(lines=1, label="Enter your prompt", interactive=True)
            with gr.Column(scale=1, min_width=60):
                generate_button = gr.Button("Generate")
            with gr.Column(scale=1, min_width=60):
                reset_button = gr.Button("Reset")

        with gr.Row():
            steps = gr.Number(value=50, precision=0, label="Steps", interactive=True)
            randomize_seed = gr.Radio(
                ["Fix Seed", "Randomize Seed"],
                label="Seed Type",
                value="Fix Seed",
                type="index",
                show_label=False,
                interactive=True,
            )
            seed = gr.Number(value=8978, precision=0, label="Seed", interactive=True)
            
        with gr.Row():
            cfg_scale = gr.Slider(
                label="Text Guidance Scale", minimum=0, maximum=10, value=3.0, step=0.1
            )
            sag_scale = gr.Slider(
                label="Self-Attention Guidance Scale", minimum=0, maximum=1.0, value=0.75, step=0.05
            )

        with gr.Row():
            ori_image = gr.Image(label="CFG", type="pil", interactive=False)
            sag_image = gr.Image(label="SAG + CFG", type="pil", interactive=False)
            ori_image.style(height=512, width=512)
            sag_image.style(height=512, width=512)

            
        ex = gr.Examples(
            examples=examples,
            fn=generate,
            inputs=[
                prompt,
                steps,
                randomize_seed,
                seed,
                cfg_scale,
                sag_scale,
            ],
            outputs=[ori_image, sag_image, seed],
            cache_examples=False,
        )

        gr.Markdown(help_text)
        
        generate_button.click(
            fn=generate,
            inputs=[
                prompt,
                steps,
                randomize_seed,
                seed,
                cfg_scale,
                sag_scale,
            ],
            outputs=[ori_image, sag_image, seed],
        )
        reset_button.click(
            fn=reset,
            inputs=[],
            outputs=[steps, randomize_seed, seed, cfg_scale, sag_scale, ori_image, sag_image],
        )

    demo.queue(concurrency_count=1)
    demo.launch(share=False)


if __name__ == "__main__":
    main()