File size: 6,524 Bytes
95b1a30
 
 
 
 
 
f043377
 
 
95b1a30
 
2153359
95b1a30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2153359
 
95b1a30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55190ff
95b1a30
 
55190ff
95b1a30
 
f043377
95b1a30
f043377
55190ff
83e23a7
95b1a30
 
 
 
 
 
 
 
 
 
 
 
55190ff
95b1a30
 
f043377
95b1a30
 
2153359
 
 
 
 
 
 
 
95b1a30
 
2153359
 
 
 
 
 
 
 
 
 
 
 
 
95b1a30
 
 
6529b84
 
95b1a30
 
 
 
 
 
 
432ca0d
95b1a30
 
 
 
 
 
 
 
f043377
95b1a30
 
 
 
 
 
 
 
f043377
95b1a30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203

from typing import Tuple

import requests
import random
import numpy as np
import gradio as gr
import spaces
import torch
from PIL import Image
from diffusers import FluxInpaintPipeline
from diffusers import FluxImg2ImgPipeline

MAX_SEED = np.iinfo(np.int32).max
IMAGE_SIZE = 1024
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"


def remove_background(image: Image.Image, threshold: int = 50) -> Image.Image:
    image = image.convert("RGBA")
    data = image.getdata()
    new_data = []
    for item in data:
        avg = sum(item[:3]) / 3
        if avg < threshold:
            new_data.append((0, 0, 0, 0))
        else:
            new_data.append(item)

    image.putdata(new_data)
    return image


pipe = FluxInpaintPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16).to(DEVICE)
pipe2 = FluxImg2ImgPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16).to(DEVICE)

def resize_image_dimensions(
    original_resolution_wh: Tuple[int, int],
    maximum_dimension: int = IMAGE_SIZE
) -> Tuple[int, int]:
    width, height = original_resolution_wh

    # if width <= maximum_dimension and height <= maximum_dimension:
    #     width = width - (width % 32)
    #     height = height - (height % 32)
    #     return width, height

    if width > height:
        scaling_factor = maximum_dimension / width
    else:
        scaling_factor = maximum_dimension / height

    new_width = int(width * scaling_factor)
    new_height = int(height * scaling_factor)

    new_width = new_width - (new_width % 32)
    new_height = new_height - (new_height % 32)

    return new_width, new_height


@spaces.GPU(duration=80)
def process(
    input_image_editor: dict,
    input_text: str,
    seed_slicer: int,
    randomize_seed_checkbox: bool,
    strength_slider: float,
    num_inference_steps_slider: int,
    progress=gr.Progress(track_tqdm=True)
):
    if not input_text:
        gr.Info("Please enter a text prompt.")
        return None, None

    image = input_image_editor['background']
    mask = input_image_editor['layers'][0]

    if not image:
        gr.Info("Please upload an image.")
        return result, None
        
    width, height = resize_image_dimensions(original_resolution_wh=image.size)
    resized_image = image.resize((width, height), Image.LANCZOS)
    if randomize_seed_checkbox:
       seed_slicer = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed_slicer)
 
    if not mask:
        gr.Info("Please draw a mask on the image.")
        pipe2.load_lora_weights("SIGMitch/KIT")
        result = pipe2(
        prompt=input_text,
        image=resized_image,
        width=width,
        height=height,
        strength=strength_slider,
        generator=generator,
        joint_attention_kwargs={"scale": 1.2},
        num_inference_steps=num_inference_steps_slider
        ).images[0]
        print('INFERENCE DONE')
        return result, None

    resized_mask = mask.resize((width, height), Image.LANCZOS)

    pipe.load_lora_weights("SIGMitch/KIT")
    result = pipe(
        prompt=input_text,
        image=resized_image,
        mask_image=resized_mask,
        width=width,
        height=height,
        strength=strength_slider,
        generator=generator,
        joint_attention_kwargs={"scale": 1.2},
        num_inference_steps=num_inference_steps_slider
    ).images[0]
    print('INFERENCE DONE')
    return result, resized_mask


with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            input_image_editor_component = gr.ImageEditor(
                label='Image',
                type='pil',
                sources=["upload", "webcam"],
                image_mode='RGB',
                layers=False,
                brush=gr.Brush(colors=["#FFFFFF"], color_mode="fixed"))

            with gr.Row():
                input_text_component = gr.Text(
                    label="Prompt",
                    show_label=False,
                    max_lines=1,
                    placeholder="Enter your prompt",
                    container=False,
                )
                submit_button_component = gr.Button(
                    value='Submit', variant='primary', scale=0)

            with gr.Accordion("Advanced Settings", open=False):
                seed_slicer_component = gr.Slider(
                    label="Seed",
                    minimum=0,
                    maximum=MAX_SEED,
                    step=1,
                    value=42,
                )

                randomize_seed_checkbox_component = gr.Checkbox(
                    label="Randomize seed", value=True)

                with gr.Row():
                    strength_slider_component = gr.Slider(
                        label="Strength",
                        info="Indicates extent to transform the reference `image`. "
                             "Must be between 0 and 1. `image` is used as a starting "
                             "point and more noise is added the higher the `strength`.",
                        minimum=0,
                        maximum=1,
                        step=0.01,
                        value=0.85,
                    )

                    num_inference_steps_slider_component = gr.Slider(
                        label="Number of inference steps",
                        info="The number of denoising steps. More denoising steps "
                             "usually lead to a higher quality image at the",
                        minimum=1,
                        maximum=50,
                        step=1,
                        value=20,
                    )
        with gr.Column():
            output_image_component = gr.Image(
                type='pil', image_mode='RGB', label='Generated image', format="png")
            with gr.Accordion("Debug", open=False):
                output_mask_component = gr.Image(
                    type='pil', image_mode='RGB', label='Input mask', format="png")

    submit_button_component.click(
        fn=process,
        inputs=[
            input_image_editor_component,
            input_text_component,
            seed_slicer_component,
            randomize_seed_checkbox_component,
            strength_slider_component,
            num_inference_steps_slider_component
        ],
        outputs=[
            output_image_component,
            output_mask_component
        ]
    )

demo.launch(debug=False, show_error=True)