Hugging Face's logo

Spaces:
hansyan
/
PeRFlow-SDXL
Runtime error

App Files Community
2
PeRFlow-SDXL
File size: 3,963 Bytes 
3aaaaf0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
# import spaces
import random
import gradio as gr
import numpy as np
import torch
from PIL import Image

def setup_seed(seed):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    torch.backends.cudnn.deterministic = True
    
if torch.cuda.is_available():
    device = "cuda:0"
else:
    device = "cpu"

### PeRFlow-T2I
from diffusers import StableDiffusionXLPipeline
pipe = StableDiffusionXLPipeline.from_pretrained("hansyan/perflow-sdxl-dreamshaper", torch_dtype=torch.float16, use_safetensors=True, variant="v0-fix")
from src.scheduler_perflow import PeRFlowScheduler
pipe.scheduler = PeRFlowScheduler.from_config(pipe.scheduler.config, prediction_type="ddim_eps", num_time_windows=4)
pipe.to("cuda:0", torch.float16)
# pipe_t2i = None


### gradio
# @spaces.GPU
def generate(text, num_inference_steps, cfg_scale, seed):
    setup_seed(int(seed))
    num_inference_steps = int(num_inference_steps)
    cfg_scale = float(cfg_scale)
    
    prompt_prefix = "photorealistic, uhd, high resolution, high quality, highly detailed; "
    neg_prompt = "distorted, blur, low-quality, haze, out of focus"
    text = prompt_prefix + text
    samples = pipe(
            prompt              = [text],
            negative_prompt     = [neg_prompt],
            height              = 1024,
            width               = 1024,
            num_inference_steps = num_inference_steps,
            guidance_scale      = cfg_scale,
            output_type         = 'pt',
        ).images
    samples = samples.squeeze(0).permute(1, 2, 0).cpu().numpy()*255.
    samples = samples.astype(np.uint8)
    samples = Image.fromarray(samples[:, :, :3])
    return samples


# layout
css = """
h1 {
    text-align: center;
    display:block;
}
h2 {
    text-align: center;
    display:block;
}
h3 {
    text-align: center;
    display:block;
}
.gradio-container {
  max-width: 768px !important;
}
"""
with gr.Blocks(title="PeRFlow-SDXL", css=css) as interface:
    gr.Markdown(
    """
    # PeRFlow-SDXL

    GitHub: [https://github.com/magic-research/piecewise-rectified-flow](https://github.com/magic-research/piecewise-rectified-flow) <br/>
    Models: [https://huggingface.co/hansyan/perflow-sdxl-dreamshaper](https://huggingface.co/hansyan/perflow-sdxl-dreamshaper) 
    
    <br/>
    """
    )
    
    with gr.Column():
        text = gr.Textbox(
            label="Input Prompt", 
            value="masterpiece, A closeup face photo of girl, wearing a rain coat, in the street, heavy rain, bokeh"
        )
        with gr.Row():
            num_inference_steps = gr.Dropdown(label='Num Inference Steps',choices=[4,5,6,7,8], value=6, interactive=True)
            cfg_scale = gr.Dropdown(label='CFG scale',choices=[1.5, 2.0, 2.5], value=2.0, interactive=True)
            seed = gr.Textbox(label="Random Seed", value=42)
            submit = gr.Button(scale=1, variant='primary')
    
    # with gr.Column():
    # with gr.Row():
    output_image = gr.Image(label='Generated Image')

    gr.Markdown(
    """
    Here are some examples provided:
    - “masterpiece, A closeup face photo of girl, wearing a rain coat, in the street, heavy rain, bokeh”
    - “RAW photo, a handsome man, wearing a black coat, outside, closeup face”
    - “RAW photo, a red luxury car, studio light”
    - “masterpiece, A beautiful cat bask in the sun”
    """
    )
    
    # activate
    text.submit(
        fn=generate,
        inputs=[text, num_inference_steps, cfg_scale, seed],
        outputs=[output_image],
    )
    seed.submit(
        fn=generate,
        inputs=[text, num_inference_steps, cfg_scale, seed],
        outputs=[output_image],
    )
    submit.click(fn=generate,
        inputs=[text, num_inference_steps, cfg_scale, seed],
        outputs=[output_image],
    )



if __name__ == '__main__':
    interface.queue(max_size=10)
    # interface.launch()
    interface.launch(share=True)