Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,150 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
|
4 |
+
import spaces
|
5 |
+
import torch
|
6 |
+
import spaces
|
7 |
+
import random
|
8 |
+
|
9 |
+
from diffusers import AutoPipelineForText2Image
|
10 |
+
from PIL import Image
|
11 |
+
|
12 |
+
|
13 |
+
MAX_SEED = np.iinfo(np.int32).max
|
14 |
+
MAX_IMAGE_SIZE = 2048
|
15 |
+
|
16 |
+
pipe = AutoPipelineForText2Image.from_pretrained(
|
17 |
+
"ostris/Flex.2-preview",
|
18 |
+
custom_pipeline="pipeline.py",
|
19 |
+
torch_dtype=torch.bfloat16,
|
20 |
+
).to("cuda")
|
21 |
+
|
22 |
+
# def calculate_optimal_dimensions(image: Image.Image):
|
23 |
+
# # Extract the original dimensions
|
24 |
+
# original_width, original_height = image.size
|
25 |
+
|
26 |
+
# # Set constants
|
27 |
+
# MIN_ASPECT_RATIO = 9 / 16
|
28 |
+
# MAX_ASPECT_RATIO = 16 / 9
|
29 |
+
# FIXED_DIMENSION = 1024
|
30 |
+
|
31 |
+
# # Calculate the aspect ratio of the original image
|
32 |
+
# original_aspect_ratio = original_width / original_height
|
33 |
+
|
34 |
+
# # Determine which dimension to fix
|
35 |
+
# if original_aspect_ratio > 1: # Wider than tall
|
36 |
+
# width = FIXED_DIMENSION
|
37 |
+
# height = round(FIXED_DIMENSION / original_aspect_ratio)
|
38 |
+
# else: # Taller than wide
|
39 |
+
# height = FIXED_DIMENSION
|
40 |
+
# width = round(FIXED_DIMENSION * original_aspect_ratio)
|
41 |
+
|
42 |
+
# # Ensure dimensions are multiples of 8
|
43 |
+
# width = (width // 8) * 8
|
44 |
+
# height = (height // 8) * 8
|
45 |
+
|
46 |
+
# # Enforce aspect ratio limits
|
47 |
+
# calculated_aspect_ratio = width / height
|
48 |
+
# if calculated_aspect_ratio > MAX_ASPECT_RATIO:
|
49 |
+
# width = (height * MAX_ASPECT_RATIO // 8) * 8
|
50 |
+
# elif calculated_aspect_ratio < MIN_ASPECT_RATIO:
|
51 |
+
# height = (width / MIN_ASPECT_RATIO // 8) * 8
|
52 |
+
|
53 |
+
# # Ensure width and height remain above the minimum dimensions
|
54 |
+
# width = max(width, 576) if width == FIXED_DIMENSION else width
|
55 |
+
# height = max(height, 576) if height == FIXED_DIMENSION else height
|
56 |
+
|
57 |
+
# return width, height
|
58 |
+
|
59 |
+
@spaces.GPU
|
60 |
+
def infer(edit_images, prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
|
61 |
+
image = edit_images["background"]
|
62 |
+
width, height = calculate_optimal_dimensions(image)
|
63 |
+
mask = edit_images["layers"][0]
|
64 |
+
if randomize_seed:
|
65 |
+
seed = random.randint(0, MAX_SEED)
|
66 |
+
image = pipe(
|
67 |
+
prompt=prompt,
|
68 |
+
# image=image,
|
69 |
+
# mask_image=mask,
|
70 |
+
inpaint_image=image,
|
71 |
+
inpaint_mask=mask,
|
72 |
+
height=height,
|
73 |
+
width=width,
|
74 |
+
guidance_scale=guidance_scale,
|
75 |
+
num_inference_steps=num_inference_steps,
|
76 |
+
generator=torch.Generator("cpu").manual_seed(seed)
|
77 |
+
).images[0]
|
78 |
+
return image, seed
|
79 |
+
|
80 |
+
examples = [
|
81 |
+
"a tiny astronaut hatching from an egg on the moon",
|
82 |
+
"a cat holding a sign that says hello world",
|
83 |
+
"an anime illustration of a wiener schnitzel",
|
84 |
+
]
|
85 |
+
|
86 |
+
css="""
|
87 |
+
#col-container {
|
88 |
+
margin: 0 auto;
|
89 |
+
max-width: 1000px;
|
90 |
+
}
|
91 |
+
"""
|
92 |
+
|
93 |
+
with gr.Blocks(css=css) as demo:
|
94 |
+
|
95 |
+
with gr.Column(elem_id="col-container"):
|
96 |
+
gr.Markdown(f"""# Flex.2 Preview - inpaint
|
97 |
+
""")
|
98 |
+
with gr.Row():
|
99 |
+
with gr.Column():
|
100 |
+
edit_image = gr.ImageEditor(
|
101 |
+
label='Upload and draw mask for inpainting',
|
102 |
+
type='pil',
|
103 |
+
sources=["upload", "webcam"],
|
104 |
+
image_mode='RGB',
|
105 |
+
layers=False,
|
106 |
+
brush=gr.Brush(colors=["#FFFFFF"], color_mode="fixed"),
|
107 |
+
height=600
|
108 |
+
)
|
109 |
+
prompt = gr.Text(
|
110 |
+
label="Prompt",
|
111 |
+
show_label=False,
|
112 |
+
max_lines=1,
|
113 |
+
placeholder="Enter your prompt",
|
114 |
+
container=False,
|
115 |
+
)
|
116 |
+
run_button = gr.Button("Run")
|
117 |
+
|
118 |
+
result = gr.Image(label="Result", show_label=False)
|
119 |
+
|
120 |
+
with gr.Accordion("Advanced Settings", open=False):
|
121 |
+
|
122 |
+
seed = gr.Slider(
|
123 |
+
label="Seed",
|
124 |
+
minimum=0,
|
125 |
+
maximum=MAX_SEED,
|
126 |
+
step=1,
|
127 |
+
value=0,
|
128 |
+
)
|
129 |
+
|
130 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
131 |
+
|
132 |
+
with gr.Row():
|
133 |
+
|
134 |
+
height = gr.Slider(64, 2048, value=512, step=64, label="Height")
|
135 |
+
width = gr.Slider(64, 2048, value=512, step=64, label="Width")
|
136 |
+
|
137 |
+
with gr.Row():
|
138 |
+
|
139 |
+
guidance_scale = gr.Slider(0.0, 20.0, value=3.5, step=0.1, label="Guidance Scale")
|
140 |
+
|
141 |
+
num_inference_steps = gr.Slider(1, 100, value=50, step=1, label="Inference Steps")
|
142 |
+
|
143 |
+
gr.on(
|
144 |
+
triggers=[run_button.click, prompt.submit],
|
145 |
+
fn = infer,
|
146 |
+
inputs = [edit_image, prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
|
147 |
+
outputs = [result, seed]
|
148 |
+
)
|
149 |
+
|
150 |
+
demo.launch()
|