File size: 5,893 Bytes
bd92452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f343415
bd92452
 
 
 
 
 
7c25d4e
bd92452
 
 
 
 
a2dbbb3
bd92452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e09b028
bd92452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
abf293f
bd92452
 
22d6257
 
 
 
 
bd92452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a2dbbb3
bd92452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a2dbbb3
bd92452
 
 
 
 
 
 
 
 
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
import gradio as gr
import spaces
import torch
from diffusers import AutoencoderKL, TCDScheduler, DPMSolverMultistepScheduler
from diffusers.models.model_loading_utils import load_state_dict
from gradio_imageslider import ImageSlider
from huggingface_hub import hf_hub_download
from PIL import ImageDraw, ImageFont, Image

from controlnet_union import ControlNetModel_Union
from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline

MODELS = {
    "RealVisXL V5.0 Lightning": "SG161222/RealVisXL_V5.0_Lightning",
}

config_file = hf_hub_download(
    "xinsir/controlnet-union-sdxl-1.0",
    filename="config_promax.json",
)

config = ControlNetModel_Union.load_config(config_file)
controlnet_model = ControlNetModel_Union.from_config(config)
model_file = hf_hub_download(
    "xinsir/controlnet-union-sdxl-1.0",
    filename="diffusion_pytorch_model_promax.safetensors",
)
state_dict = load_state_dict(model_file)
model, _, _, _, _ = ControlNetModel_Union._load_pretrained_model(
    controlnet_model, state_dict, model_file, "xinsir/controlnet-union-sdxl-1.0"
)
model.to(device="cuda", dtype=torch.float16)

vae = AutoencoderKL.from_pretrained(
    "madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
).to("cuda")

pipe = StableDiffusionXLFillPipeline.from_pretrained(
    "SG161222/RealVisXL_V5.0_Lightning",
    torch_dtype=torch.float16,
    vae=vae,
    controlnet=model,
    variant="fp16",
).to("cuda")

pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')


@spaces.GPU
def fill_image(prompt, negative_prompt, image, model_selection, paste_back, guidance_scale, num_steps):
    (
        prompt_embeds,
        negative_prompt_embeds,
        pooled_prompt_embeds,
        negative_pooled_prompt_embeds,
    ) = pipe.encode_prompt(prompt, "cuda", True,negative_prompt=negative_prompt)

    source = image["background"]
    mask = image["layers"][0]

    alpha_channel = mask.split()[3]
    binary_mask = alpha_channel.point(lambda p: p > 0 and 255)
    cnet_image = source.copy()
    cnet_image.paste(0, (0, 0), binary_mask)

    for image in pipe(
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        pooled_prompt_embeds=pooled_prompt_embeds,
        negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        image=cnet_image,
        guidance_scale = guidance_scale,
        num_inference_steps = num_steps,
    ):
        yield image, cnet_image

    print(f"{model_selection=}")
    print(f"{paste_back=}")

    if paste_back:
        image = image.convert("RGBA")
        cnet_image.paste(image, (0, 0), binary_mask)
    else:
        cnet_image = image

    yield source, cnet_image


def clear_result():
    return gr.update(value=None)


title = """<h1 align="center">Inpaint</h1>"""

with gr.Blocks() as demo:
    gr.HTML(title)
    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(
                label="Prompt",
                info="Describe what to inpaint the mask with",
                lines=3,
            )
            
            with gr.Accordion("Advanced Options", open=False):
                negative_prompt = gr.Textbox(
                    label="Negative Prompt",
                    info="Describe what you dont want in the mask",
                    lines=3,
                )
                guidance_scale = gr.Slider(
                    minimum=1,
                    maximum=10,
                    value=1.5,
                    step=0.1,
                    label="Guidance Scale"
                )
                num_steps = gr.Slider(
                    minimum=5,
                    maximum=100,
                    value=10,
                    step=1,
                    label="Steps"
                )
            
            input_image = gr.ImageMask(
                type="pil", label="Input Image",crop_size=(1200,1200), layers=False
            )
        with gr.Column():
            model_selection = gr.Dropdown(
                choices=list(MODELS.keys()),
                value="RealVisXL V5.0 Lightning",
                label="Model",
            )

            with gr.Row():
                with gr.Column():
                    run_button = gr.Button("Generate")

                with gr.Column():
                    paste_back = gr.Checkbox(True, label="Paste back original")

            result = ImageSlider(
                interactive=False,
                label="Generated Image",
                type="pil"
            )

    use_as_input_button = gr.Button("Use as Input Image", visible=False)

    def use_output_as_input(output_image):
        return gr.update(value=output_image[1])

    use_as_input_button.click(
        fn=use_output_as_input, inputs=[result], outputs=[input_image]
    )

    run_button.click(
        fn=clear_result,
        inputs=None,
        outputs=result,
    ).then(
        fn=lambda: gr.update(visible=False),
        inputs=None,
        outputs=use_as_input_button,
    ).then(
        fn=fill_image,
        inputs=[prompt, negative_prompt, input_image, model_selection, paste_back, guidance_scale, num_steps],
        outputs=result,
    ).then(
        fn=lambda: gr.update(visible=True),
        inputs=None,
        outputs=use_as_input_button,
    )

    prompt.submit(
        fn=clear_result,
        inputs=None,
        outputs=result,
    ).then(
        fn=lambda: gr.update(visible=False),
        inputs=None,
        outputs=use_as_input_button,
    ).then(
        fn=fill_image,
        inputs=[prompt, negative_prompt, input_image, model_selection, paste_back, guidance_scale, num_steps],
        outputs=result,
    ).then(
        fn=lambda: gr.update(visible=True),
        inputs=None,
        outputs=use_as_input_button,
    )


demo.queue(max_size=12).launch(share=False)