File size: 14,663 Bytes
edb0494
6405936
 
 
 
 
 
edb0494
6405936
 
edb0494
a7d8817
d49f90c
a7d8817
6405936
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ef457d
 
 
 
 
 
 
 
aee2c4c
 
4ab7724
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aee2c4c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4ab7724
8ef457d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4ab7724
 
 
 
 
a7d8817
8ef457d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4ab7724
a7d8817
 
6405936
8ef457d
4ab7724
 
 
 
 
 
 
 
6405936
 
 
 
 
 
4ab7724
6405936
4ab7724
6405936
 
a7d8817
6405936
a7d8817
15a8627
8ef457d
 
 
 
aeb7d74
8ef457d
5bc9409
 
4b78a6c
5bc9409
 
4b78a6c
5bc9409
 
aee2c4c
 
 
 
5bc9409
aeb7d74
6405936
fb5d273
 
 
 
 
aee2c4c
 
fb5d273
 
6405936
aee2c4c
 
 
34b60b0
 
 
 
 
 
 
97567b1
 
40a5fd5
97567b1
 
 
9cdaf5d
 
4a91cdc
 
 
 
 
 
 
6405936
 
97567b1
976671e
 
 
 
 
 
 
8ef457d
976671e
8ef457d
 
 
 
 
 
 
4ab7724
40a5fd5
8ef457d
aee2c4c
8ef457d
 
40a5fd5
 
8ef457d
 
 
 
 
4ab7724
5bc9409
8ef457d
5bc9409
 
 
 
8ef457d
5bc9409
 
 
 
 
 
8ef457d
5bc9409
d03bc23
5bc9409
 
8ef457d
 
 
 
 
 
 
5bc9409
aee2c4c
 
 
 
 
 
 
 
 
 
 
 
 
 
8ef457d
c837d9c
4ab7724
8ef457d
 
 
 
c837d9c
8ef457d
c837d9c
4ab7724
976671e
 
 
 
 
8ef457d
 
34b60b0
 
8ef457d
 
 
97567b1
8ef457d
 
 
 
 
 
5bc9409
8ef457d
aeb7d74
8ef457d
 
5bc9409
8ef457d
fb5d273
aee2c4c
 
 
 
fb5d273
 
aeb7d74
aee2c4c
 
 
 
 
 
 
 
 
 
 
aeb7d74
 
34b60b0
6405936
 
 
34b60b0
976671e
8ef457d
aee2c4c
6405936
34b60b0
 
 
 
 
8ef457d
 
 
6405936
 
34b60b0
4ab7724
 
 
34b60b0
4ab7724
8ef457d
aee2c4c
4ab7724
34b60b0
 
 
 
 
8ef457d
 
 
4ab7724
6405936
8ef457d
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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
import gradio as gr
import spaces
import torch
from diffusers import AutoencoderKL, TCDScheduler
from diffusers.models.model_loading_utils import load_state_dict
from gradio_imageslider import ImageSlider
from huggingface_hub import hf_hub_download

from controlnet_union import ControlNetModel_Union
from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline

from PIL import Image, ImageDraw
import numpy as np

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)


def can_expand(source_width, source_height, target_width, target_height, alignment):
    """Checks if the image can be expanded based on the alignment."""
    if alignment in ("Left", "Right") and source_width >= target_width:
        return False
    if alignment in ("Top", "Bottom") and source_height >= target_height:
        return False
    return True

@spaces.GPU(duration=24)
def infer(image, width, height, overlap_width, num_inference_steps, resize_option, custom_resize_size, prompt_input=None, alignment="Middle"):
    source = image
    target_size = (width, height)
    overlap = overlap_width

    # Upscale if source is smaller than target in both dimensions
    if source.width < target_size[0] and source.height < target_size[1]:
        scale_factor = min(target_size[0] / source.width, target_size[1] / source.height)
        new_width = int(source.width * scale_factor)
        new_height = int(source.height * scale_factor)
        source = source.resize((new_width, new_height), Image.LANCZOS)

    if source.width > target_size[0] or source.height > target_size[1]:
        scale_factor = min(target_size[0] / source.width, target_size[1] / source.height)
        new_width = int(source.width * scale_factor)
        new_height = int(source.height * scale_factor)
        source = source.resize((new_width, new_height), Image.LANCZOS)
    
    if resize_option == "Full":
        resize_size = max(source.width, source.height)
    elif resize_option == "1/2":
        resize_size = max(source.width, source.height) // 2
    elif resize_option == "1/3":
        resize_size = max(source.width, source.height) // 3
    elif resize_option == "1/4":
        resize_size = max(source.width, source.height) // 4
    else:  # Custom
        resize_size = custom_resize_size

    aspect_ratio = source.height / source.width
    new_width = resize_size
    new_height = int(resize_size * aspect_ratio)
    source = source.resize((new_width, new_height), Image.LANCZOS)

    if not can_expand(source.width, source.height, target_size[0], target_size[1], alignment):
        alignment = "Middle"

    # Calculate margins based on alignment
    if alignment == "Middle":
        margin_x = (target_size[0] - source.width) // 2
        margin_y = (target_size[1] - source.height) // 2
    elif alignment == "Left":
        margin_x = 0
        margin_y = (target_size[1] - source.height) // 2
    elif alignment == "Right":
        margin_x = target_size[0] - source.width
        margin_y = (target_size[1] - source.height) // 2
    elif alignment == "Top":
        margin_x = (target_size[0] - source.width) // 2
        margin_y = 0
    elif alignment == "Bottom":
        margin_x = (target_size[0] - source.width) // 2
        margin_y = target_size[1] - source.height

    background = Image.new('RGB', target_size, (255, 255, 255))
    background.paste(source, (margin_x, margin_y))

    mask = Image.new('L', target_size, 255)
    mask_draw = ImageDraw.Draw(mask)

    # Adjust mask generation based on alignment
    if alignment == "Middle":
        mask_draw.rectangle([
            (margin_x + overlap, margin_y + overlap),
            (margin_x + source.width - overlap, margin_y + source.height - overlap)
        ], fill=0)
    elif alignment == "Left":
        mask_draw.rectangle([
            (margin_x, margin_y),
            (margin_x + source.width - overlap, margin_y + source.height)
        ], fill=0)
    elif alignment == "Right":
        mask_draw.rectangle([
            (margin_x + overlap, margin_y),
            (margin_x + source.width, margin_y + source.height)
        ], fill=0)
    elif alignment == "Top":
        mask_draw.rectangle([
            (margin_x, margin_y),
            (margin_x + source.width, margin_y + source.height - overlap)
        ], fill=0)
    elif alignment == "Bottom":
        mask_draw.rectangle([
            (margin_x, margin_y + overlap),
            (margin_x + source.width, margin_y + source.height)
        ], fill=0)

    cnet_image = background.copy()
    cnet_image.paste(0, (0, 0), mask)

    final_prompt = f"{prompt_input} , high quality, 4k"

    (
        prompt_embeds,
        negative_prompt_embeds,
        pooled_prompt_embeds,
        negative_pooled_prompt_embeds,
    ) = pipe.encode_prompt(final_prompt, "cuda", True)

    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,
        num_inference_steps=num_inference_steps
    ):
        yield cnet_image, image

    image = image.convert("RGBA")
    cnet_image.paste(image, (0, 0), mask)

    yield background, cnet_image

def clear_result():
    """Clears the result ImageSlider."""
    return gr.update(value=None)

def preload_presets(target_ratio, ui_width, ui_height):
    """Updates the width and height sliders based on the selected aspect ratio."""
    if target_ratio == "9:16":
        changed_width = 720
        changed_height = 1280
        return changed_width, changed_height, gr.update(open=False)
    elif target_ratio == "16:9":
        changed_width = 1280
        changed_height = 720
        return changed_width, changed_height, gr.update(open=False)
    elif target_ratio == "1:1":
        changed_width = 1024
        changed_height = 1024
        return changed_width, changed_height, gr.update(open=False)
    elif target_ratio == "Custom":
        return ui_width, ui_height, gr.update(open=True)

def select_the_right_preset(user_width, user_height):
    if user_width == 720 and user_height == 1280:
        return "9:16"
    elif user_width == 1280 and user_height == 720:
        return "16:9"
    elif user_width == 1024 and user_height == 1024:
        return "1:1"
    else:
        return "Custom"

def toggle_custom_resize_slider(resize_option):
    return gr.update(visible=(resize_option == "Custom"))

def update_history(new_image, history):
    """Updates the history gallery with the new image."""
    if history is None:
        history = []
    history.insert(0, new_image)
    return history

css = """
.gradio-container {
    width: 1200px !important;
}
"""

title = """<h1 align="center">Diffusers Image Outpaint</h1>
<div align="center">Drop an image you would like to extend, pick your expected ratio and hit Generate.</div>
<div style="display: flex; justify-content: center; align-items: center; text-align: center;">
    <p style="display: flex;gap: 6px;">
         <a href="https://huggingface.co/spaces/fffiloni/diffusers-image-outpout?duplicate=true">
            <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-md.svg" alt="Duplicate this Space">
        </a> to skip the queue and enjoy faster inference on the GPU of your choice 
    </p>
</div>
"""

with gr.Blocks(css=css) as demo:
    with gr.Column():
        gr.HTML(title)

        with gr.Row():
            with gr.Column():
                input_image = gr.Image(
                    type="pil",
                    label="Input Image"
                )

                with gr.Row():
                    with gr.Column(scale=2):
                        prompt_input = gr.Textbox(label="Prompt (Optional)")
                    with gr.Column(scale=1):
                        run_button = gr.Button("Generate")

                with gr.Row():
                    target_ratio = gr.Radio(
                        label="Expected Ratio",
                        choices=["9:16", "16:9", "1:1", "Custom"],
                        value="9:16",
                        scale=2
                    )
                    
                    alignment_dropdown = gr.Dropdown(
                        choices=["Middle", "Left", "Right", "Top", "Bottom"],
                        value="Middle",
                        label="Alignment"
                    )

                with gr.Accordion(label="Advanced settings", open=False) as settings_panel:
                    with gr.Column():
                        with gr.Row():
                            width_slider = gr.Slider(
                                label="Width",
                                minimum=720,
                                maximum=1536,
                                step=8,
                                value=720,  # Set a default value
                            )
                            height_slider = gr.Slider(
                                label="Height",
                                minimum=720,
                                maximum=1536,
                                step=8,
                                value=1280,  # Set a default value
                            )
                        with gr.Row():
                            num_inference_steps = gr.Slider(label="Steps", minimum=4, maximum=12, step=1, value=8)
                            overlap_width = gr.Slider(
                                label="Mask overlap width",
                                minimum=1,
                                maximum=50,
                                value=42,
                                step=1
                            )
                        with gr.Row():
                            resize_option = gr.Radio(
                                label="Resize input image",
                                choices=["Full", "1/2", "1/3", "1/4", "Custom"],
                                value="Full"
                            )
                            custom_resize_size = gr.Slider(
                                label="Custom resize size",
                                minimum=64,
                                maximum=1024,
                                step=8,
                                value=512,
                                visible=False
                            )
                            
                gr.Examples(
                    examples=[
                        ["./examples/example_1.webp", 1280, 720, "Middle"],
                        ["./examples/example_2.jpg", 1440, 810, "Left"],
                        ["./examples/example_3.jpg", 1024, 1024, "Top"],
                        ["./examples/example_3.jpg", 1024, 1024, "Bottom"],
                    ],
                    inputs=[input_image, width_slider, height_slider, alignment_dropdown],
                )

            with gr.Column():
                result = ImageSlider(
                    interactive=False,
                    label="Generated Image",
                )
                use_as_input_button = gr.Button("Use as Input Image", visible=False)

        history_gallery = gr.Gallery(label="History", columns=6, height=240, object_fit="contain")

    def use_output_as_input(output_image):
        """Sets the generated output as the new input image."""
        return gr.update(value=output_image[1])

    use_as_input_button.click(
        fn=use_output_as_input,
        inputs=[result],
        outputs=[input_image]
    )
    
    target_ratio.change(
        fn=preload_presets,
        inputs=[target_ratio, width_slider, height_slider],
        outputs=[width_slider, height_slider, settings_panel],
        queue=False
    )

    width_slider.change(
        fn=select_the_right_preset,
        inputs=[width_slider, height_slider],
        outputs=[target_ratio],
        queue=False
    )

    height_slider.change(
        fn=select_the_right_preset,
        inputs=[width_slider, height_slider],
        outputs=[target_ratio],
        queue=False
    )

    resize_option.change(
        fn=toggle_custom_resize_slider,
        inputs=[resize_option],
        outputs=[custom_resize_size],
        queue=False
    )
    
    run_button.click(  # Clear the result
        fn=clear_result,
        inputs=None,
        outputs=result,
    ).then(  # Generate the new image
        fn=infer,
        inputs=[input_image, width_slider, height_slider, overlap_width, num_inference_steps,
                resize_option, custom_resize_size, prompt_input, alignment_dropdown],
        outputs=result,
    ).then(  # Update the history gallery
        fn=lambda x, history: update_history(x[1], history),
        inputs=[result, history_gallery],
        outputs=history_gallery,
    ).then(  # Show the "Use as Input Image" button
        fn=lambda: gr.update(visible=True),
        inputs=None,
        outputs=use_as_input_button,
    )

    prompt_input.submit(  # Clear the result
        fn=clear_result,
        inputs=None,
        outputs=result,
    ).then(  # Generate the new image
        fn=infer,
        inputs=[input_image, width_slider, height_slider, overlap_width, num_inference_steps,
                resize_option, custom_resize_size, prompt_input, alignment_dropdown],
        outputs=result,
    ).then(  # Update the history gallery
        fn=lambda x, history: update_history(x[1], history),
        inputs=[result, history_gallery],
        outputs=history_gallery,
    ).then(  # Show the "Use as Input Image" button
        fn=lambda: gr.update(visible=True),
        inputs=None,
        outputs=use_as_input_button,
    )

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