File size: 4,224 Bytes
2a35740
8bb91d1
 
 
ff99100
2a35740
88adfd9
8bb91d1
88adfd9
 
 
 
 
 
8bb91d1
88adfd9
 
 
265d8de
8bb91d1
 
 
 
2a35740
d315a84
ff99100
88adfd9
 
 
 
243b399
88adfd9
2a35740
 
8bb91d1
 
 
 
2a35740
8bb91d1
 
 
 
 
 
 
 
 
 
 
2a35740
 
ff99100
2a35740
 
 
 
265d8de
2a35740
8bb91d1
 
88adfd9
 
8bb91d1
2a35740
8bb91d1
 
2a35740
 
8bb91d1
2a35740
88adfd9
2a35740
 
 
 
8bb91d1
 
 
2a35740
88adfd9
2a35740
d315a84
2a35740
88adfd9
8bb91d1
2a35740
88adfd9
 
 
 
2a35740
88adfd9
2a35740
 
 
 
 
 
 
 
 
75d29b6
 
 
2a35740
 
 
75d29b6
 
 
2a35740
 
 
818a2f4
75d29b6
 
2a35740
 
 
75d29b6
 
795ac1a
75d29b6
 
 
 
 
8bb91d1
 
 
d315a84
8bb91d1
 
 
2a35740
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
import spaces
import gradio as gr
import numpy as np
import torch
from PIL import Image
from diffusers import DDPMScheduler, StableDiffusionPipeline, DDIMScheduler, UNet2DConditionModel
from diffusers import StableDiffusionInstructPix2PixPipeline, LCMScheduler

# InstructPix2Pix with LCM specified scheduler
pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(
       "timbrooks/instruct-pix2pix", torch_dtype=torch.float16
       )
pipe = pipe.to("cuda")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# Adapt the InstructPix2Pix model using the LoRA parameters
adapter_id = "latent-consistency/lcm-lora-sdv1-5"
pipe.load_lora_weights(adapter_id)
pipe.to('cuda')

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

@spaces.GPU(duration=30)
def infer(image, edit_instruction, guidance_scale, n_steps):
    image = Image.fromarray(image).resize((512, 512))
    image = pipe(prompt=edit_instruction, 
             image=image,
             num_inference_steps=n_steps, 
             guidance_scale=guidance_scale,
             image_guidance_scale=1.0
             ).images[0]

    return image

css="""
#col-container {
    margin: 0 auto;
    max-width: 1024px;
}
"""

if torch.cuda.is_available():
    power_device = "GPU"
else:
    power_device = "CPU"

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(
            f"""
        # ⚡ Instruct-pix2pix with Consistency Distillation⚡ 
        Currently running on {power_device}
        """
        )
        gr.Markdown(
            "If you enjoy the space, feel free to give a ⭐ to the <a href='https://github.com/yandex-research/invertible-cd' target='_blank'>Github Repo</a>. [![GitHub Stars](https://img.shields.io/github/stars/quickjkee/instruct-pix2pix-distill?style=social)](https://github.com/quickjkee/instruct-pix2pix-distill)"
        )
        with gr.Row():
            
            edit_instruction = gr.Text(
                label="Edit instruction",
                max_lines=1,
                placeholder="Enter your prompt",
            )
            
        
        with gr.Row():
            
            with gr.Column():
                image = gr.Image(label="Input image", height=512, width=512, show_label=False)
            with gr.Column():
                result = gr.Image(label="Result", height=512, width=512, show_label=False)

        with gr.Accordion("Advanced Settings", open=True):
            
            with gr.Row():
                
                guidance_scale = gr.Slider(
                    label="guidance scale",
                    minimum=1.0,
                    maximum=5.0,
                    step=1.0,
                    value=2.0,
                )

                n_steps = gr.Slider(
                    label="inference steps",
                    minimum=1.0,
                    maximum=10.0,
                    step=1.0,
                    value=4.0,
                )

        with gr.Row():
            run_button = gr.Button("Edit", scale=0)

        with gr.Row():
            examples = [
                [
                    "examples/orig_3.jpg", #input_image
                    "turn apples into oranges", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
                [
                    "examples/orig_1.jpg", #input_image
                    "Make it a Modigliani painting", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
                [
                    "examples/orig_2.jpg", #input_image
                    "Turn a teddy bear into panda", #tgt_prompt
                    2, #guidance_scale
                    4
                ],
            ]
  
            gr.Examples(
               examples = examples,
               inputs =[image, edit_instruction, guidance_scale, n_steps],
               outputs=[
                        result
                        ],
               fn=infer, cache_examples=True
            )

    run_button.click(
        fn = infer,
        inputs=[image, edit_instruction, guidance_scale, n_steps],
        outputs = [result]
    )

demo.queue().launch()