File size: 11,678 Bytes
5d92a23
 
1a21d80
5d92a23
 
0d89801
5d92a23
 
 
0d89801
9ed6403
 
a715048
7231bf4
a715048
6d9ca04
 
 
d718ec3
 
 
 
0d89801
 
7756cc0
0d89801
 
5d92a23
1dcf1d9
f0e8d1f
f37c54f
63c0f18
03400c5
1a21d80
6d9ca04
a715048
 
 
 
 
 
 
 
f56aaf9
a715048
 
 
 
 
 
f56aaf9
a715048
 
 
 
f56aaf9
 
a715048
 
44b4b3d
 
 
 
 
 
 
 
 
 
 
 
 
 
6d9ca04
 
 
 
 
 
 
 
 
0d89801
 
5d92a23
 
1dcf1d9
5d92a23
 
 
1dcf1d9
 
 
 
11414b7
5d92a23
 
 
 
 
 
 
 
d2b30ac
 
5d92a23
 
 
 
1a21d80
5d92a23
 
d718ec3
 
a4143d8
 
2008ad3
 
66437b5
 
5d92a23
 
 
 
 
 
 
0d89801
 
1a21d80
0d89801
d718ec3
0d89801
5d92a23
0d89801
d718ec3
 
 
 
 
 
7a8b6d6
d718ec3
 
 
 
 
 
 
0d89801
 
1a21d80
0d89801
5d92a23
0d89801
1a21d80
a4143d8
 
0d89801
a715048
 
 
 
2008ad3
a715048
7a8b6d6
a715048
 
5d92a23
 
 
a715048
 
7a8b6d6
a715048
 
 
 
 
 
 
 
 
6d9ca04
a715048
 
 
7231bf4
0d89801
 
 
 
 
 
 
 
 
 
 
 
3e075bb
d718ec3
 
 
 
 
 
 
 
 
 
 
 
 
66437b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2008ad3
66437b5
8ca9ae0
2008ad3
 
8ca9ae0
 
6d9ca04
2008ad3
66437b5
2008ad3
 
 
66437b5
6d9ca04
2008ad3
66437b5
 
2008ad3
66437b5
 
 
8ca9ae0
5d92a23
2008ad3
9ed6403
66437b5
 
5d92a23
 
 
 
 
11414b7
5d92a23
 
 
1dcf1d9
5d92a23
a4143d8
 
 
 
 
 
 
 
 
5d92a23
 
 
1dcf1d9
 
 
5d92a23
 
 
1dcf1d9
5d92a23
 
 
 
1dcf1d9
 
5d92a23
 
 
6d9ca04
5d92a23
0d89801
7231bf4
 
5d92a23
 
1a21d80
0d89801
 
 
 
 
d718ec3
 
a4143d8
 
2008ad3
 
66437b5
 
5d92a23
 
 
 
 
0d89801
 
5d92a23
 
0d89801
 
 
7ba0702
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
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 huggingface_hub import login
import os
import time
from gradio_imageslider import ImageSlider

from diffusers import FlowMatchEulerDiscreteScheduler, AutoencoderKL
from diffusers.models.transformers.transformer_flux import FluxTransformer2DModel
from transformers import CLIPTextModel, CLIPTokenizer,T5EncoderModel, T5TokenizerFast
import requests
from io import BytesIO
import PIL.Image
import requests

MARKDOWN = """
# FLUX.1 Inpainting with lora
"""

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

#login(token=HF_TOKEN)

bfl_repo="black-forest-labs/FLUX.1-dev"

class calculateDuration:
    def __init__(self, activity_name=""):
        self.activity_name = activity_name

    def __enter__(self):
        self.start_time = time.time()
        self.start_time_formatted = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(self.start_time))
        print(f"Activity: {self.activity_name}, Start time: {self.start_time_formatted}")
        return self
    
    def __exit__(self, exc_type, exc_value, traceback):
        self.end_time = time.time()
        self.elapsed_time = self.end_time - self.start_time
        self.end_time_formatted = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(self.end_time))
        
        if self.activity_name:
            print(f"Elapsed time for {self.activity_name}: {self.elapsed_time:.6f} seconds")
        else:
            print(f"Elapsed time: {self.elapsed_time:.6f} seconds")
        
        print(f"Activity: {self.activity_name}, End time: {self.start_time_formatted}")


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

# text_encoder = CLIPTextModel.from_pretrained(os.path.join(os.getcwd(), "flux_text_encoders/clip_l.safetensors"), torch_dtype=dtype)
# tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14", torch_dtype=dtype)
# text_encoder_2 = T5EncoderModel.from_pretrained(os.path.join(os.getcwd(), "flux_text_encoders/t5xxl_fp8_e4m3fn.safetensors"), torch_dtype=dtype)
# tokenizer_2 = T5TokenizerFast.from_pretrained(bfl_repo, subfolder="tokenizer_2", torch_dtype=dtype)
# vae = AutoencoderKL.from_pretrained(bfl_repo, subfolder="vae", torch_dtype=dtype)
# transformer = FluxTransformer2DModel.from_pretrained(bfl_repo, subfolder="transformer", torch_dtype=dtype)


pipe = FluxInpaintPipeline.from_pretrained(bfl_repo, 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=100)
def process(
    input_image_editor: dict,
    image_url: str,
    mask_url: str,
    blur_mask: bool,
    blur_factor: int,
    lora_path: str,
    lora_weights: str,
    lora_scale: float,
    trigger_word: str,
    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

    # default image edtiro
    image = input_image_editor['background']
    mask = input_image_editor['layers'][0]

    if image_url:
        print("start to fetch image from url", image_url)
        response = requests.get(image_url)
        response.raise_for_status()
        image = PIL.Image.open(BytesIO(response.content))
        print("fetch image success")

    if mask_url:
        print("start to fetch mask from url", mask_url)
        response = requests.get(mask_url)
        response.raise_for_status()
        mask = PIL.Image.open(BytesIO(response.content))
        print("fetch mask success")

    if not image:
        gr.Info("Please upload an image.")
        return None, None

    if not mask:
        gr.Info("Please draw a mask on the image.")
        return None, None
    if blur_mask:
        mask = pipe.mask_processor.blur(mask, blur_factor=blur_factor)

    with calculateDuration("resize image"):
        width, height = resize_image_dimensions(original_resolution_wh=image.size)
        resized_image = image.resize((width, height), Image.LANCZOS)
        resized_mask = mask.resize((width, height), Image.LANCZOS)
    
    with calculateDuration("load lora"):
        print(lora_path, lora_weights)
        pipe.load_lora_weights(lora_path, weight_name=lora_weights)
    
    if randomize_seed_checkbox:
        seed_slicer = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed_slicer)

    with calculateDuration("run pipe"):
        print(input_text, width, height, strength_slider, num_inference_steps_slider, lora_scale)
        result = pipe(
            prompt=f"{input_text} {trigger_word}",
            image=resized_image,
            mask_image=resized_mask,
            width=width,
            height=height,
            strength=strength_slider,
            generator=generator,
            num_inference_steps=num_inference_steps_slider,
            max_sequence_length=256,
            joint_attention_kwargs={"scale": lora_scale},
        ).images[0]
    
    return [resized_image, result], resized_mask


with gr.Blocks() as demo:
    gr.Markdown(MARKDOWN)
    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"))
                
            image_url =  gr.Textbox(
                    label="image url",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your image url (Optional)",
                )
            mask_url =  gr.Textbox(
                    label="Mask image url",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your mask image url (Optional)",
                )
        
            with gr.Accordion("Prompt Settings", open=True):

                input_text_component = gr.Textbox(
                    label="Inpaint prompt",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your prompt",
                )
                trigger_word = gr.Textbox(
                    label="Lora trigger word",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your lora trigger word here",
                    value="a photo of TOK"
                    
                )

                submit_button_component = gr.Button(
                    value='Submit', variant='primary', scale=0)

            with gr.Accordion("Lora Settings", open=True):
                lora_path = gr.Textbox(
                    label="Lora model path",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your model path",
                    info="Currently, only LoRA hosted on Hugging Face'model can be loaded properly.",
                    value="XLabs-AI/flux-RealismLora"
                )
                lora_weights = gr.Textbox(
                    label="Lora weights",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter your lora weights name",
                    value="lora.safetensors"
                )
                lora_scale = gr.Slider(
                    label="Lora scale",
                    show_label=True,
                    minimum=0,
                    maximum=1,
                    step=0.1,
                    value=0.9,
                )
                
            with gr.Accordion("Advanced Settings", open=True):
                
                
                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)

                blur_mask = gr.Checkbox(
                    label="if blur mask", value=False)
                blur_factor =  gr.Slider(
                    label="blur factor",
                    minimum=0,
                    maximum=50,
                    step=1,
                    value=33,
                )
                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=28,
                    )
        with gr.Column():
            output_image_component = ImageSlider(label="Generate image", type="pil", slider_color="pink")
            
            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,
            image_url,
            mask_url,
            blur_mask,
            blur_factor,
            lora_path,
            lora_weights,
            lora_scale,
            trigger_word,
            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, share=False)