Spaces:
Running
Running
Upload 38 files
Browse files- .gitattributes +17 -0
- README.md +3 -4
- app.py +382 -0
- assets/No-Image-Placeholder.png +0 -0
- assets/examples/1-newton.jpg +0 -0
- assets/examples/1-output-1.png +3 -0
- assets/examples/1-output-2.png +3 -0
- assets/examples/1-output-3.png +3 -0
- assets/examples/1-style-1.jpg +0 -0
- assets/examples/2-output-1.png +3 -0
- assets/examples/2-output-2.png +3 -0
- assets/examples/2-output-3.png +3 -0
- assets/examples/2-output-4.png +3 -0
- assets/examples/2-style-1.jpg +0 -0
- assets/examples/2-stylegan2-ffhq-0100.png +3 -0
- assets/examples/2-stylegan2-ffhq-0293.png +3 -0
- assets/examples/3-output-1.png +3 -0
- assets/examples/3-output-2.png +3 -0
- assets/examples/3-output-3.png +3 -0
- assets/examples/3-output-4.png +3 -0
- assets/examples/3-style-1.png +3 -0
- assets/examples/3-style-2.jpg +0 -0
- assets/examples/3-style-3.jpg +0 -0
- assets/examples/3-stylegan2-ffhq-0293.png +3 -0
- assets/examples/3-stylegan2-ffhq-0381.png +3 -0
- assets/examples/Biden-1.jpg +0 -0
- assets/examples/Biden-2.jpg +0 -0
- assets/examples/Biden-3.jpg +0 -0
- assets/examples/Biden-4.jpg +0 -0
- assets/examples/Trump-1.jpg +0 -0
- assets/examples/Trump-2.jpg +0 -0
- assets/examples/Trump-3.jpg +0 -0
- assets/examples/Trump-4.jpg +3 -0
- requirements.txt +1 -0
- src/__init__.py +0 -0
- src/generation.py +79 -0
- src/log.py +18 -0
- src/process.py +277 -0
- src/util.py +58 -0
.gitattributes
CHANGED
@@ -33,3 +33,20 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
assets/examples/1-output-1.png filter=lfs diff=lfs merge=lfs -text
|
37 |
+
assets/examples/1-output-2.png filter=lfs diff=lfs merge=lfs -text
|
38 |
+
assets/examples/1-output-3.png filter=lfs diff=lfs merge=lfs -text
|
39 |
+
assets/examples/2-output-1.png filter=lfs diff=lfs merge=lfs -text
|
40 |
+
assets/examples/2-output-2.png filter=lfs diff=lfs merge=lfs -text
|
41 |
+
assets/examples/2-output-3.png filter=lfs diff=lfs merge=lfs -text
|
42 |
+
assets/examples/2-output-4.png filter=lfs diff=lfs merge=lfs -text
|
43 |
+
assets/examples/2-stylegan2-ffhq-0100.png filter=lfs diff=lfs merge=lfs -text
|
44 |
+
assets/examples/2-stylegan2-ffhq-0293.png filter=lfs diff=lfs merge=lfs -text
|
45 |
+
assets/examples/3-output-1.png filter=lfs diff=lfs merge=lfs -text
|
46 |
+
assets/examples/3-output-2.png filter=lfs diff=lfs merge=lfs -text
|
47 |
+
assets/examples/3-output-3.png filter=lfs diff=lfs merge=lfs -text
|
48 |
+
assets/examples/3-output-4.png filter=lfs diff=lfs merge=lfs -text
|
49 |
+
assets/examples/3-style-1.png filter=lfs diff=lfs merge=lfs -text
|
50 |
+
assets/examples/3-stylegan2-ffhq-0293.png filter=lfs diff=lfs merge=lfs -text
|
51 |
+
assets/examples/3-stylegan2-ffhq-0381.png filter=lfs diff=lfs merge=lfs -text
|
52 |
+
assets/examples/Trump-4.jpg filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
@@ -1,13 +1,12 @@
|
|
1 |
---
|
2 |
title: UniPortrait
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
colorTo: red
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 4.
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: apache-2.0
|
11 |
---
|
12 |
-
|
13 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
title: UniPortrait
|
3 |
+
emoji: 👩🎨
|
4 |
+
colorFrom: purple
|
5 |
colorTo: red
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 4.31.2
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: apache-2.0
|
11 |
---
|
|
|
12 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
app.py
ADDED
@@ -0,0 +1,382 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
##!/usr/bin/python3
|
2 |
+
# -*- coding: utf-8 -*-
|
3 |
+
# @Time : 2024-07-31
|
4 |
+
# @Author : Junjie He
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
from src.process import (
|
8 |
+
text_to_single_id_generation_process,
|
9 |
+
text_to_multi_id_generation_process,
|
10 |
+
image_to_single_id_generation_process,
|
11 |
+
)
|
12 |
+
|
13 |
+
|
14 |
+
def text_to_single_id_generation_block():
|
15 |
+
gr.Markdown("## Text-to-Single-ID Generation")
|
16 |
+
gr.HTML(text_to_single_id_description)
|
17 |
+
gr.HTML(text_to_single_id_tips)
|
18 |
+
with gr.Row():
|
19 |
+
with gr.Column(scale=1, min_width=100):
|
20 |
+
prompt = gr.Textbox(value="", label='Prompt', lines=2)
|
21 |
+
negative_prompt = gr.Textbox(value="nsfw", label='Negative Prompt')
|
22 |
+
image_resolution = gr.Dropdown(choices=["768x512", "512x512", "512x768"], value="512x512",
|
23 |
+
label="Image Resolution (HxW)")
|
24 |
+
run_button = gr.Button(value="Run")
|
25 |
+
|
26 |
+
with gr.Accordion("Advanced Options", open=True):
|
27 |
+
seed = gr.Slider(label="Seed (-1 indicates random)", minimum=-1, maximum=2147483647, step=1, value=-1)
|
28 |
+
|
29 |
+
faceid_scale = gr.Slider(label="Face ID Scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
30 |
+
face_structure_scale = gr.Slider(label="Face Structure Scale", minimum=0.0, maximum=1.0,
|
31 |
+
step=0.01, value=0.1)
|
32 |
+
|
33 |
+
style_scale = gr.Slider(label="style_scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
34 |
+
|
35 |
+
use_sr = gr.Checkbox(label="RealESRGAN 2x", value=True)
|
36 |
+
|
37 |
+
with gr.Column(scale=3, min_width=100):
|
38 |
+
with gr.Row(equal_height=False):
|
39 |
+
pil_faceid = gr.Image(type="pil", label="ID Image")
|
40 |
+
with gr.Accordion("ID Supplements", open=True):
|
41 |
+
with gr.Row():
|
42 |
+
pil_supp_faceids = gr.File(file_count="multiple", file_types=["image"],
|
43 |
+
type="filepath", label="Additional ID Images")
|
44 |
+
with gr.Row():
|
45 |
+
with gr.Column(scale=1, min_width=100):
|
46 |
+
pil_mix_faceid_1 = gr.Image(type="pil", label="Mix ID 1")
|
47 |
+
mix_scale_1 = gr.Slider(label="Mix Scale 1", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
|
48 |
+
with gr.Column(scale=1, min_width=100):
|
49 |
+
pil_mix_faceid_2 = gr.Image(type="pil", label="Mix ID 2")
|
50 |
+
mix_scale_2 = gr.Slider(label="Mix Scale 2", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
|
51 |
+
pil_style = gr.Image(type="pil", label="Style")
|
52 |
+
|
53 |
+
with gr.Row():
|
54 |
+
example_output = gr.Image(type="pil", label="(Example Output)", visible=False)
|
55 |
+
result_gallery = gr.Gallery(label='Output', show_label=True, elem_id="gallery", columns=4, preview=True,
|
56 |
+
format="png")
|
57 |
+
with gr.Row():
|
58 |
+
examples = [
|
59 |
+
[
|
60 |
+
"A young man with short black hair, wearing a black hoodie with a hood, was paired with a blue denim jacket with yellow details.",
|
61 |
+
"assets/examples/1-newton.jpg",
|
62 |
+
"assets/No-Image-Placeholder.png",
|
63 |
+
"assets/examples/1-output-1.png",
|
64 |
+
],
|
65 |
+
[
|
66 |
+
"A young man with short black hair, wearing a black hoodie with a hood, was paired with a blue denim jacket with yellow details.",
|
67 |
+
"assets/examples/1-newton.jpg",
|
68 |
+
"assets/examples/1-style-1.jpg",
|
69 |
+
"assets/examples/1-output-2.png",
|
70 |
+
],
|
71 |
+
]
|
72 |
+
gr.Examples(
|
73 |
+
label="Examples",
|
74 |
+
examples=examples,
|
75 |
+
inputs=[prompt, pil_faceid, pil_style, example_output],
|
76 |
+
)
|
77 |
+
ips = [
|
78 |
+
pil_faceid, pil_supp_faceids,
|
79 |
+
pil_mix_faceid_1, mix_scale_1,
|
80 |
+
pil_mix_faceid_2, mix_scale_2,
|
81 |
+
faceid_scale, face_structure_scale,
|
82 |
+
prompt, negative_prompt,
|
83 |
+
pil_style, style_scale,
|
84 |
+
seed, image_resolution, use_sr,
|
85 |
+
]
|
86 |
+
run_button.click(fn=text_to_single_id_generation_process, inputs=ips, outputs=[result_gallery])
|
87 |
+
|
88 |
+
|
89 |
+
def text_to_multi_id_generation_block():
|
90 |
+
gr.Markdown("## Text-to-Multi-ID Generation")
|
91 |
+
gr.HTML(text_to_multi_id_description)
|
92 |
+
gr.HTML(text_to_multi_id_tips)
|
93 |
+
with gr.Row():
|
94 |
+
with gr.Column(scale=1, min_width=100):
|
95 |
+
prompt = gr.Textbox(value="", label='Prompt', lines=2)
|
96 |
+
negative_prompt = gr.Textbox(value="nsfw", label='Negative Prompt')
|
97 |
+
image_resolution = gr.Dropdown(choices=["768x512", "512x512", "512x768"], value="512x512",
|
98 |
+
label="Image Resolution (HxW)")
|
99 |
+
run_button = gr.Button(value="Run")
|
100 |
+
|
101 |
+
with gr.Accordion("Advanced Options", open=True):
|
102 |
+
seed = gr.Slider(label="Seed (-1 indicates random)", minimum=-1, maximum=2147483647, step=1, value=-1)
|
103 |
+
|
104 |
+
faceid_scale = gr.Slider(label="Face ID Scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
105 |
+
face_structure_scale = gr.Slider(label="Face Structure Scale", minimum=0.0, maximum=1.0,
|
106 |
+
step=0.01, value=0.3)
|
107 |
+
style_scale = gr.Slider(label="style_scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
108 |
+
|
109 |
+
use_sr = gr.Checkbox(label="RealESRGAN 2x", value=True)
|
110 |
+
|
111 |
+
with gr.Column(scale=3, min_width=100):
|
112 |
+
with gr.Row(equal_height=False):
|
113 |
+
with gr.Column(scale=1, min_width=100):
|
114 |
+
pil_faceid_1st = gr.Image(type="pil", label="First ID")
|
115 |
+
with gr.Accordion("First ID Supplements", open=False):
|
116 |
+
with gr.Row():
|
117 |
+
pil_supp_faceids_1st = gr.File(file_count="multiple", file_types=["image"],
|
118 |
+
type="filepath", label="Additional ID Images")
|
119 |
+
with gr.Row():
|
120 |
+
with gr.Column(scale=1, min_width=100):
|
121 |
+
pil_mix_faceid_1_1st = gr.Image(type="pil", label="Mix ID 1")
|
122 |
+
mix_scale_1_1st = gr.Slider(label="Mix Scale 1", minimum=0.0, maximum=1.0, step=0.01,
|
123 |
+
value=0.0)
|
124 |
+
with gr.Column(scale=1, min_width=100):
|
125 |
+
pil_mix_faceid_2_1st = gr.Image(type="pil", label="Mix ID 2")
|
126 |
+
mix_scale_2_1st = gr.Slider(label="Mix Scale 2", minimum=0.0, maximum=1.0, step=0.01,
|
127 |
+
value=0.0)
|
128 |
+
with gr.Column(scale=1, min_width=100):
|
129 |
+
pil_faceid_2nd = gr.Image(type="pil", label="Second ID")
|
130 |
+
with gr.Accordion("Second ID Supplements", open=False):
|
131 |
+
with gr.Row():
|
132 |
+
pil_supp_faceids_2nd = gr.File(file_count="multiple", file_types=["image"],
|
133 |
+
type="filepath", label="Additional ID Images")
|
134 |
+
with gr.Row():
|
135 |
+
with gr.Column(scale=1, min_width=100):
|
136 |
+
pil_mix_faceid_1_2nd = gr.Image(type="pil", label="Mix ID 1")
|
137 |
+
mix_scale_1_2nd = gr.Slider(label="Mix Scale 1", minimum=0.0, maximum=1.0, step=0.01,
|
138 |
+
value=0.0)
|
139 |
+
with gr.Column(scale=1, min_width=100):
|
140 |
+
pil_mix_faceid_2_2nd = gr.Image(type="pil", label="Mix ID 2")
|
141 |
+
mix_scale_2_2nd = gr.Slider(label="Mix Scale 2", minimum=0.0, maximum=1.0, step=0.01,
|
142 |
+
value=0.0)
|
143 |
+
with gr.Column(scale=1, min_width=100):
|
144 |
+
pil_style = gr.Image(type="pil", label="Style")
|
145 |
+
|
146 |
+
with gr.Row():
|
147 |
+
example_output = gr.Image(type="pil", label="(Example Output)", visible=False)
|
148 |
+
result_gallery = gr.Gallery(label='Output', show_label=True, elem_id="gallery", columns=4, preview=True,
|
149 |
+
format="png")
|
150 |
+
with gr.Row():
|
151 |
+
examples = [
|
152 |
+
[
|
153 |
+
"两个女人在欢笑和快乐中被捕捉到,他们的脸上洋溢着真挚的幸福,背景是日落时分的宁静海滩。这幅画以柔和的风格描绘,捕捉了这一刻的温暖和宁静。",
|
154 |
+
"assets/examples/2-stylegan2-ffhq-0100.png",
|
155 |
+
"assets/examples/2-stylegan2-ffhq-0293.png",
|
156 |
+
"assets/No-Image-Placeholder.png",
|
157 |
+
"assets/examples/2-output-1.png",
|
158 |
+
],
|
159 |
+
[
|
160 |
+
"The two female models are drinking coffee. The background was off-white.",
|
161 |
+
"assets/examples/2-stylegan2-ffhq-0100.png",
|
162 |
+
"assets/examples/2-stylegan2-ffhq-0293.png",
|
163 |
+
"assets/examples/2-style-1.jpg",
|
164 |
+
"assets/examples/2-output-2.png",
|
165 |
+
],
|
166 |
+
]
|
167 |
+
gr.Examples(
|
168 |
+
label="Examples",
|
169 |
+
examples=examples,
|
170 |
+
inputs=[prompt, pil_faceid_1st, pil_faceid_2nd, pil_style, example_output],
|
171 |
+
)
|
172 |
+
with gr.Row():
|
173 |
+
examples = [
|
174 |
+
[
|
175 |
+
"Two men in an American poster.",
|
176 |
+
"assets/examples/Trump-1.jpg",
|
177 |
+
["assets/examples/Trump-2.jpg", "assets/examples/Trump-3.jpg", "assets/examples/Trump-4.jpg"],
|
178 |
+
"assets/examples/Biden-1.jpg",
|
179 |
+
["assets/examples/Biden-2.jpg", "assets/examples/Biden-3.jpg", "assets/examples/Biden-4.jpg"],
|
180 |
+
"assets/examples/2-output-4.png",
|
181 |
+
],
|
182 |
+
[
|
183 |
+
"Two men engaged in a vigorous handshake, both wearing expressions of enthusiasm and determination, set against a backdrop of a bustling business district. The image is crafted in a sleek and modern digital art style, conveying the dynamic and competitive nature of their interaction.",
|
184 |
+
"assets/examples/Trump-1.jpg",
|
185 |
+
["assets/examples/Trump-2.jpg", "assets/examples/Trump-3.jpg", "assets/examples/Trump-4.jpg"],
|
186 |
+
"assets/examples/Biden-1.jpg",
|
187 |
+
["assets/examples/Biden-2.jpg", "assets/examples/Biden-3.jpg", "assets/examples/Biden-4.jpg"],
|
188 |
+
"assets/examples/2-output-3.png",
|
189 |
+
],
|
190 |
+
]
|
191 |
+
gr.Examples(
|
192 |
+
label="Examples (Multiple References)",
|
193 |
+
examples=examples,
|
194 |
+
inputs=[prompt, pil_faceid_1st, pil_supp_faceids_1st, pil_faceid_2nd, pil_supp_faceids_2nd, example_output],
|
195 |
+
)
|
196 |
+
ips = [
|
197 |
+
pil_faceid_1st, pil_supp_faceids_1st,
|
198 |
+
pil_mix_faceid_1_1st, mix_scale_1_1st,
|
199 |
+
pil_mix_faceid_2_1st, mix_scale_2_1st,
|
200 |
+
pil_faceid_2nd, pil_supp_faceids_2nd,
|
201 |
+
pil_mix_faceid_1_2nd, mix_scale_1_2nd,
|
202 |
+
pil_mix_faceid_2_2nd, mix_scale_2_2nd,
|
203 |
+
faceid_scale, face_structure_scale,
|
204 |
+
prompt, negative_prompt,
|
205 |
+
pil_style, style_scale,
|
206 |
+
seed, image_resolution, use_sr,
|
207 |
+
]
|
208 |
+
run_button.click(fn=text_to_multi_id_generation_process, inputs=ips, outputs=[result_gallery])
|
209 |
+
|
210 |
+
|
211 |
+
def image_to_single_id_generation_block():
|
212 |
+
gr.Markdown("## Image-to-Single-ID Generation")
|
213 |
+
gr.HTML(image_to_single_id_description)
|
214 |
+
gr.HTML(image_to_single_id_tips)
|
215 |
+
with gr.Row():
|
216 |
+
with gr.Column(scale=1, min_width=100):
|
217 |
+
image_resolution = gr.Dropdown(choices=["768x512", "512x512", "512x768"], value="512x512",
|
218 |
+
label="Image Resolution (HxW)")
|
219 |
+
run_button = gr.Button(value="Run")
|
220 |
+
|
221 |
+
with gr.Accordion("Advanced Options", open=True):
|
222 |
+
seed = gr.Slider(label="Seed (-1 indicates random)", minimum=-1, maximum=2147483647, step=1, value=-1)
|
223 |
+
|
224 |
+
style_scale = gr.Slider(label="Reference Scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
225 |
+
faceid_scale = gr.Slider(label="Face ID Scale", minimum=0.0, maximum=1.0, step=0.01, value=0.7)
|
226 |
+
face_structure_scale = gr.Slider(label="Face Structure Scale", minimum=0.0, maximum=1.0, step=0.01,
|
227 |
+
value=0.3)
|
228 |
+
|
229 |
+
use_sr = gr.Checkbox(label="RealESRGAN 2x", value=True)
|
230 |
+
|
231 |
+
with gr.Column(scale=3, min_width=100):
|
232 |
+
with gr.Row(equal_height=False):
|
233 |
+
pil_style = gr.Image(type="pil", label="Portrait Reference")
|
234 |
+
pil_faceid = gr.Image(type="pil", label="ID Image")
|
235 |
+
with gr.Accordion("ID Supplements", open=True):
|
236 |
+
with gr.Row():
|
237 |
+
pil_supp_faceids = gr.File(file_count="multiple", file_types=["image"],
|
238 |
+
type="filepath", label="Additional ID Images")
|
239 |
+
with gr.Row():
|
240 |
+
with gr.Column(scale=1, min_width=100):
|
241 |
+
pil_mix_faceid_1 = gr.Image(type="pil", label="Mix ID 1")
|
242 |
+
mix_scale_1 = gr.Slider(label="Mix Scale 1", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
|
243 |
+
with gr.Column(scale=1, min_width=100):
|
244 |
+
pil_mix_faceid_2 = gr.Image(type="pil", label="Mix ID 2")
|
245 |
+
mix_scale_2 = gr.Slider(label="Mix Scale 2", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
|
246 |
+
with gr.Row():
|
247 |
+
with gr.Column(scale=3, min_width=100):
|
248 |
+
example_output = gr.Image(type="pil", label="(Example Output)", visible=False)
|
249 |
+
result_gallery = gr.Gallery(label='Output', show_label=True, elem_id="gallery", columns=4,
|
250 |
+
preview=True, format="png")
|
251 |
+
with gr.Row():
|
252 |
+
examples = [
|
253 |
+
[
|
254 |
+
"assets/examples/3-style-1.png",
|
255 |
+
"assets/examples/3-stylegan2-ffhq-0293.png",
|
256 |
+
0.7,
|
257 |
+
0.3,
|
258 |
+
"assets/examples/3-output-1.png",
|
259 |
+
],
|
260 |
+
[
|
261 |
+
"assets/examples/3-style-1.png",
|
262 |
+
"assets/examples/3-stylegan2-ffhq-0293.png",
|
263 |
+
0.6,
|
264 |
+
0.0,
|
265 |
+
"assets/examples/3-output-2.png",
|
266 |
+
],
|
267 |
+
[
|
268 |
+
"assets/examples/3-style-2.jpg",
|
269 |
+
"assets/examples/3-stylegan2-ffhq-0381.png",
|
270 |
+
0.7,
|
271 |
+
0.3,
|
272 |
+
"assets/examples/3-output-3.png",
|
273 |
+
],
|
274 |
+
[
|
275 |
+
"assets/examples/3-style-3.jpg",
|
276 |
+
"assets/examples/3-stylegan2-ffhq-0381.png",
|
277 |
+
0.6,
|
278 |
+
0.0,
|
279 |
+
"assets/examples/3-output-4.png",
|
280 |
+
],
|
281 |
+
]
|
282 |
+
gr.Examples(
|
283 |
+
label="Examples",
|
284 |
+
examples=examples,
|
285 |
+
inputs=[pil_style, pil_faceid, faceid_scale, face_structure_scale, example_output],
|
286 |
+
)
|
287 |
+
ips = [
|
288 |
+
pil_faceid, pil_supp_faceids,
|
289 |
+
pil_mix_faceid_1, mix_scale_1,
|
290 |
+
pil_mix_faceid_2, mix_scale_2,
|
291 |
+
faceid_scale, face_structure_scale,
|
292 |
+
pil_style, style_scale,
|
293 |
+
seed, image_resolution, use_sr,
|
294 |
+
]
|
295 |
+
run_button.click(fn=image_to_single_id_generation_process, inputs=ips, outputs=[result_gallery])
|
296 |
+
|
297 |
+
|
298 |
+
if __name__ == "__main__":
|
299 |
+
title = r"""
|
300 |
+
<div style="text-align: center;">
|
301 |
+
<h1> UniPortrait: A Unified Framework for Identity-Preserving Single- and Multi-Human Image Personalization </h1>
|
302 |
+
<div style="display: flex; justify-content: center; align-items: center; text-align: center;">
|
303 |
+
<a href="https://arxiv.org/pdf/xxxx.xxxxx"><img src="https://img.shields.io/badge/arXiv-xxxx.xxxxx-red"></a>
|
304 |
+
|
305 |
+
<a href='https://aigcdesigngroup.github.io/UniPortrait-Page/'><img src='https://img.shields.io/badge/Project_Page-UniPortrait-green' alt='Project Page'></a>
|
306 |
+
|
307 |
+
<a href="https://github.com/junjiehe96/UniPortrait"><img src="https://img.shields.io/badge/Github-Code-blue"></a>
|
308 |
+
</div>
|
309 |
+
</br>
|
310 |
+
</div>
|
311 |
+
"""
|
312 |
+
|
313 |
+
title_description = r"""
|
314 |
+
This is the <b>official 🤗 Gradio demo</b> for <a href='https://arxiv.org/pdf/xxxx.xxxxx' target='_blank'><b>UniPortrait: A Unified Framework for Identity-Preserving Single- and Multi-Human Image Personalization</b></a>.<br>
|
315 |
+
The demo provides three capabilities: text-to-single-ID personalization, text-to-multi-ID personalization, and image-to-single-ID personalization. All of these are based on the Stable Diffusion v1-5 model. Feel free to give them a try! 😊
|
316 |
+
"""
|
317 |
+
|
318 |
+
text_to_single_id_description = r"""🚀🚀🚀Quick start:<br>
|
319 |
+
1. Enter a text prompt (Chinese or English), Upload an image with a face, and Click the <b>Run</b> button.<br>
|
320 |
+
2. (Optional) You can also upload an image as the style reference for the results. 🤗<br>
|
321 |
+
"""
|
322 |
+
|
323 |
+
text_to_single_id_tips = r"""💡💡💡Tips:<br>
|
324 |
+
1. Try to avoid creating too small faces, as this may lead to some artifacts. (Currently, the short side length of the generated image is limited to 512)<br>
|
325 |
+
2. It's a good idea to upload multiple reference photos of your face to improve the prompt and ID consistency. Additional references can be uploaded in the "ID supplements".<br>
|
326 |
+
3. The appropriate values of "Face ID Scale" and "Face Structure Scale" are important for balancing the ID and text alignment. We recommend using "Face ID Scale" (0.5~0.7) and "Face Structure Scale" (0.0~0.4).<br>
|
327 |
+
"""
|
328 |
+
|
329 |
+
text_to_multi_id_description = r"""🚀🚀🚀Quick start:<br>
|
330 |
+
1. Enter a text prompt (Chinese or English), Upload an image with a face in "First ID" and "Second ID" blocks respectively, and Click the <b>Run</b> button.<br>
|
331 |
+
2. (Optional) You can also upload an image as the style reference for the results. 🤗<br>
|
332 |
+
"""
|
333 |
+
|
334 |
+
text_to_multi_id_tips = r"""💡💡💡Tips:<br>
|
335 |
+
1. Try to avoid creating too small faces, as this may lead to some artifacts. (Currently, the short side length of the generated image is limited to 512)<br>
|
336 |
+
2. It's a good idea to upload multiple reference photos of your face to improve the prompt and ID consistency. Additional references can be uploaded in the "ID supplements".<br>
|
337 |
+
3. The appropriate values of "Face ID Scale" and "Face Structure Scale" are important for balancing the ID and text alignment. We recommend using "Face ID Scale" (0.3~0.7) and "Face Structure Scale" (0.0~0.4).<br>
|
338 |
+
"""
|
339 |
+
|
340 |
+
image_to_single_id_description = r"""🚀🚀🚀Quick start: Upload an image as the portrait reference (can be any style), Upload a face image, and Click the <b>Run</b> button. 🤗<br>"""
|
341 |
+
|
342 |
+
image_to_single_id_tips = r"""💡💡💡Tips:<br>
|
343 |
+
1. Try to avoid creating too small faces, as this may lead to some artifacts. (Currently, the short side length of the generated image is limited to 512)<br>
|
344 |
+
2. It's a good idea to upload multiple reference photos of your face to improve ID consistency. Additional references can be uploaded in the "ID supplements".<br>
|
345 |
+
3. The appropriate values of "Face ID Scale" and "Face Structure Scale" are important for balancing the portrait reference and ID alignment. We recommend using "Face ID Scale" (0.5~0.7) and "Face Structure Scale" (0.0~0.4).<br>
|
346 |
+
"""
|
347 |
+
|
348 |
+
citation = r"""
|
349 |
+
---
|
350 |
+
📝 **Citation**
|
351 |
+
<br>
|
352 |
+
If our work is helpful for your research or applications, please cite us via:
|
353 |
+
```bibtex
|
354 |
+
@article{wang2024instantid,
|
355 |
+
title={InstantID: Zero-shot Identity-Preserving Generation in Seconds},
|
356 |
+
author={Wang, Qixun and Bai, Xu and Wang, Haofan and Qin, Zekui and Chen, Anthony},
|
357 |
+
journal={arXiv preprint arXiv:2401.07519},
|
358 |
+
year={2024}
|
359 |
+
}
|
360 |
+
```
|
361 |
+
📧 **Contact**
|
362 |
+
<br>
|
363 |
+
If you have any questions, please feel free to open an issue or directly reach us out at <b>he_junjie@zju.edu.cn</b>.
|
364 |
+
"""
|
365 |
+
|
366 |
+
block = gr.Blocks(title="UniPortrait").queue()
|
367 |
+
with block:
|
368 |
+
gr.HTML(title)
|
369 |
+
gr.HTML(title_description)
|
370 |
+
|
371 |
+
with gr.TabItem("Text-to-Single-ID"):
|
372 |
+
text_to_single_id_generation_block()
|
373 |
+
|
374 |
+
with gr.TabItem("Text-to-Multi-ID"):
|
375 |
+
text_to_multi_id_generation_block()
|
376 |
+
|
377 |
+
with gr.TabItem("Image-to-Single-ID (Stylization)"):
|
378 |
+
image_to_single_id_generation_block()
|
379 |
+
|
380 |
+
block.launch(share=True)
|
381 |
+
# block.launch(server_name='0.0.0.0', share=False, server_port=9999, allowed_paths=["/"])
|
382 |
+
# block.launch(server_name='127.0.0.1', share=False, server_port=9999, allowed_paths=["/"])
|
assets/No-Image-Placeholder.png
ADDED
assets/examples/1-newton.jpg
ADDED
assets/examples/1-output-1.png
ADDED
Git LFS Details
|
assets/examples/1-output-2.png
ADDED
Git LFS Details
|
assets/examples/1-output-3.png
ADDED
Git LFS Details
|
assets/examples/1-style-1.jpg
ADDED
assets/examples/2-output-1.png
ADDED
Git LFS Details
|
assets/examples/2-output-2.png
ADDED
Git LFS Details
|
assets/examples/2-output-3.png
ADDED
Git LFS Details
|
assets/examples/2-output-4.png
ADDED
Git LFS Details
|
assets/examples/2-style-1.jpg
ADDED
assets/examples/2-stylegan2-ffhq-0100.png
ADDED
Git LFS Details
|
assets/examples/2-stylegan2-ffhq-0293.png
ADDED
Git LFS Details
|
assets/examples/3-output-1.png
ADDED
Git LFS Details
|
assets/examples/3-output-2.png
ADDED
Git LFS Details
|
assets/examples/3-output-3.png
ADDED
Git LFS Details
|
assets/examples/3-output-4.png
ADDED
Git LFS Details
|
assets/examples/3-style-1.png
ADDED
Git LFS Details
|
assets/examples/3-style-2.jpg
ADDED
assets/examples/3-style-3.jpg
ADDED
assets/examples/3-stylegan2-ffhq-0293.png
ADDED
Git LFS Details
|
assets/examples/3-stylegan2-ffhq-0381.png
ADDED
Git LFS Details
|
assets/examples/Biden-1.jpg
ADDED
assets/examples/Biden-2.jpg
ADDED
assets/examples/Biden-3.jpg
ADDED
assets/examples/Biden-4.jpg
ADDED
assets/examples/Trump-1.jpg
ADDED
assets/examples/Trump-2.jpg
ADDED
assets/examples/Trump-3.jpg
ADDED
assets/examples/Trump-4.jpg
ADDED
Git LFS Details
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
oss2
|
src/__init__.py
ADDED
File without changes
|
src/generation.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
import time
|
4 |
+
|
5 |
+
import gradio as gr
|
6 |
+
import requests
|
7 |
+
|
8 |
+
from src.log import logger
|
9 |
+
from src.util import download_images
|
10 |
+
|
11 |
+
|
12 |
+
def call_generation(data):
|
13 |
+
url_task = os.getenv("URL_TASK")
|
14 |
+
api_key = os.getenv("API_KEY_GENERATION")
|
15 |
+
model_id = os.getenv("MODEL_ID")
|
16 |
+
url_query = os.getenv("URL_QUERY")
|
17 |
+
|
18 |
+
batch_size = 2
|
19 |
+
repeat_times = 2
|
20 |
+
|
21 |
+
headers = {
|
22 |
+
"Content-Type": "application/json",
|
23 |
+
"Accept": "application/json",
|
24 |
+
"Authorization": f"Bearer {api_key}",
|
25 |
+
"X-DashScope-Async": "enable",
|
26 |
+
}
|
27 |
+
data["model"] = model_id
|
28 |
+
data["parameters"]["n"] = batch_size
|
29 |
+
|
30 |
+
all_res_ = []
|
31 |
+
for i in range(repeat_times):
|
32 |
+
if data["parameters"]["seed"] != -1:
|
33 |
+
data["parameters"]["seed"] = data["parameters"]["seed"] * (i+1)
|
34 |
+
res_ = requests.post(url_task, data=json.dumps(data), headers=headers)
|
35 |
+
all_res_.append(res_)
|
36 |
+
|
37 |
+
all_image_data = []
|
38 |
+
for res_ in all_res_:
|
39 |
+
respose_code = res_.status_code
|
40 |
+
if 200 == respose_code:
|
41 |
+
res = json.loads(res_.content.decode())
|
42 |
+
task_id = res['output']['task_id']
|
43 |
+
logger.info(f"task_id: {task_id}: Create request success. Params: {data}")
|
44 |
+
|
45 |
+
# Async query
|
46 |
+
is_running = True
|
47 |
+
while is_running:
|
48 |
+
res_ = requests.post(f'{url_query}/{task_id}', headers=headers)
|
49 |
+
respose_code = res_.status_code
|
50 |
+
if 200 == respose_code:
|
51 |
+
res = json.loads(res_.content.decode())
|
52 |
+
if "SUCCEEDED" == res['output']['task_status']:
|
53 |
+
logger.info(f"task_id: {task_id}: Generation task query success.")
|
54 |
+
results = res['output']['results']
|
55 |
+
img_urls = [x['url'] for x in results]
|
56 |
+
logger.info(f"task_id: {task_id}: {res}")
|
57 |
+
break
|
58 |
+
elif "FAILED" != res['output']['task_status']:
|
59 |
+
logger.debug(f"task_id: {task_id}: query result...")
|
60 |
+
time.sleep(1)
|
61 |
+
else:
|
62 |
+
raise gr.Error(
|
63 |
+
"Fail to get results from Generation task. Make sure all the ID images have a clear face. If it still doesn't work, you can contact us or open an issue.")
|
64 |
+
|
65 |
+
else:
|
66 |
+
logger.error(f'task_id: {task_id}: Fail to query task result: {res_.content}')
|
67 |
+
raise gr.Error("Fail to query task result.")
|
68 |
+
|
69 |
+
logger.info(f"task_id: {task_id}: download generated images.")
|
70 |
+
img_data = download_images(img_urls, batch_size)
|
71 |
+
logger.info(f"task_id: {task_id}: Generate done.")
|
72 |
+
all_image_data += img_data
|
73 |
+
else:
|
74 |
+
logger.error(f'Fail to create Generation task: {res_.content}')
|
75 |
+
raise gr.Error("Fail to create Generation task.")
|
76 |
+
|
77 |
+
if len(all_image_data) != repeat_times * batch_size:
|
78 |
+
raise gr.Error("Fail to Generation.")
|
79 |
+
return all_image_data
|
src/log.py
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import logging
|
2 |
+
import os
|
3 |
+
from logging.handlers import RotatingFileHandler
|
4 |
+
|
5 |
+
log_file_name = "workdir/UniPortrait.log"
|
6 |
+
os.makedirs(os.path.dirname(log_file_name), exist_ok=True)
|
7 |
+
|
8 |
+
format = '[%(levelname)s] %(asctime)s "%(filename)s", line %(lineno)d, %(message)s'
|
9 |
+
logging.basicConfig(
|
10 |
+
format=format,
|
11 |
+
datefmt="%Y-%m-%d %H:%M:%S",
|
12 |
+
level=logging.INFO)
|
13 |
+
logger = logging.getLogger(name="UniPortrait-Studio")
|
14 |
+
|
15 |
+
fh = RotatingFileHandler(log_file_name, maxBytes=20000000, backupCount=3)
|
16 |
+
formatter = logging.Formatter(format, datefmt="%Y-%m-%d %H:%M:%S")
|
17 |
+
fh.setFormatter(formatter)
|
18 |
+
logger.addHandler(fh)
|
src/process.py
ADDED
@@ -0,0 +1,277 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import time
|
2 |
+
import uuid
|
3 |
+
|
4 |
+
import gradio as gr
|
5 |
+
import numpy as np
|
6 |
+
from PIL import Image
|
7 |
+
|
8 |
+
from .generation import call_generation
|
9 |
+
from .util import upload_np_2_oss
|
10 |
+
|
11 |
+
|
12 |
+
def get_image_url(request_id, pil_image, suffix):
|
13 |
+
np_image = np.array(pil_image.convert("RGB"))
|
14 |
+
image_url = upload_np_2_oss(np_image, request_id + suffix)
|
15 |
+
return image_url
|
16 |
+
|
17 |
+
|
18 |
+
def is_no_image_placeholder(pil_image):
|
19 |
+
from PIL import Image, ImageChops
|
20 |
+
no_image_placeholder = Image.open("assets/No-Image-Placeholder.png").convert("RGB")
|
21 |
+
diff = ImageChops.difference(pil_image, no_image_placeholder)
|
22 |
+
if diff.getbbox() is None:
|
23 |
+
return True
|
24 |
+
else:
|
25 |
+
return False
|
26 |
+
|
27 |
+
|
28 |
+
def text_to_single_id_generation_process(
|
29 |
+
pil_faceid=None, pil_supp_faceids=None,
|
30 |
+
pil_mix_faceid_1=None, mix_scale_1=0.0,
|
31 |
+
pil_mix_faceid_2=None, mix_scale_2=0.0,
|
32 |
+
faceid_scale=0.0, face_structure_scale=0.0,
|
33 |
+
prompt="", negative_prompt="",
|
34 |
+
pil_style=None, style_scale=0.0,
|
35 |
+
seed=-1, image_resolution="512x512", use_sr=True,
|
36 |
+
):
|
37 |
+
request_id = time.strftime('%Y%m%d-', time.localtime(time.time())) + str(uuid.uuid4())
|
38 |
+
|
39 |
+
if prompt == "":
|
40 |
+
raise gr.Error("Please enter the prompt")
|
41 |
+
|
42 |
+
if pil_style and is_no_image_placeholder(pil_style):
|
43 |
+
pil_style = None
|
44 |
+
|
45 |
+
faceid_url = []
|
46 |
+
if pil_faceid:
|
47 |
+
faceid_url.append(get_image_url(request_id, pil_faceid, suffix="_faceid.png"))
|
48 |
+
if pil_supp_faceids and len(pil_supp_faceids) > 0:
|
49 |
+
for idx, pil_supp_faceid in enumerate(pil_supp_faceids):
|
50 |
+
pil_supp_faceid = Image.open(pil_supp_faceid)
|
51 |
+
faceid_url.append(get_image_url(request_id, pil_supp_faceid, suffix=f"_supp_faceid_{idx}.png"))
|
52 |
+
|
53 |
+
mix_face_url = []
|
54 |
+
mix_scale = []
|
55 |
+
if pil_mix_faceid_1 is not None:
|
56 |
+
mix_face_url.append(
|
57 |
+
get_image_url(request_id, pil_mix_faceid_1, suffix=f"_mix_faceid_1_{mix_scale_1:.2f}.png")
|
58 |
+
)
|
59 |
+
mix_scale.append(mix_scale_1)
|
60 |
+
|
61 |
+
if pil_mix_faceid_2 is not None:
|
62 |
+
mix_face_url.append(
|
63 |
+
get_image_url(request_id, pil_mix_faceid_2, suffix=f"_mix_faceid_2_{mix_scale_2:.2f}.png")
|
64 |
+
)
|
65 |
+
mix_scale.append(mix_scale_2)
|
66 |
+
|
67 |
+
# The outer parenthesis represent the ID, and the inner parenthesis represent the mix face image for each ID.
|
68 |
+
faceid_image_url, mix_faceid_image_url, mix_faceid_scale = [], [], []
|
69 |
+
if len(faceid_url) > 0:
|
70 |
+
faceid_image_url.append(faceid_url)
|
71 |
+
mix_faceid_image_url.append(mix_face_url)
|
72 |
+
mix_faceid_scale.append(mix_scale)
|
73 |
+
|
74 |
+
if pil_style is None:
|
75 |
+
style_image_url = ""
|
76 |
+
else:
|
77 |
+
style_image_url = get_image_url(request_id, pil_style, suffix="_style.png")
|
78 |
+
|
79 |
+
data = dict(
|
80 |
+
input=dict(
|
81 |
+
faceid_image_url=faceid_image_url,
|
82 |
+
mix_faceid_image_url=mix_faceid_image_url,
|
83 |
+
mix_faceid_scale=mix_faceid_scale,
|
84 |
+
style_image_url=style_image_url,
|
85 |
+
prompt=prompt,
|
86 |
+
negative_prompt=negative_prompt
|
87 |
+
),
|
88 |
+
parameters=dict(
|
89 |
+
image_resolution=image_resolution,
|
90 |
+
faceid_scale=faceid_scale * 10.,
|
91 |
+
face_structure_scale=face_structure_scale * 10.,
|
92 |
+
style_scale=style_scale * 10.,
|
93 |
+
use_sr=int(use_sr),
|
94 |
+
seed=seed,
|
95 |
+
mode="text-to-image",
|
96 |
+
)
|
97 |
+
)
|
98 |
+
|
99 |
+
res = call_generation(data)
|
100 |
+
|
101 |
+
return res
|
102 |
+
|
103 |
+
|
104 |
+
def text_to_multi_id_generation_process(
|
105 |
+
pil_faceid_1st=None, pil_supp_faceids_1st=None,
|
106 |
+
pil_mix_faceid_1_1st=None, mix_scale_1_1st=0.0,
|
107 |
+
pil_mix_faceid_2_1st=None, mix_scale_2_1st=0.0,
|
108 |
+
pil_faceid_2nd=None, pil_supp_faceids_2nd=None,
|
109 |
+
pil_mix_faceid_1_2nd=None, mix_scale_1_2nd=0.0,
|
110 |
+
pil_mix_faceid_2_2nd=None, mix_scale_2_2nd=0.0,
|
111 |
+
faceid_scale=0.0, face_structure_scale=0.0,
|
112 |
+
prompt="", negative_prompt="",
|
113 |
+
pil_style=None, style_scale=0.0,
|
114 |
+
seed=-1, image_resolution="512x512", use_sr=True,
|
115 |
+
):
|
116 |
+
request_id = time.strftime('%Y%m%d-', time.localtime(time.time())) + str(uuid.uuid4())
|
117 |
+
|
118 |
+
if prompt == "":
|
119 |
+
raise gr.Error("Please enter the prompt")
|
120 |
+
|
121 |
+
if pil_style and is_no_image_placeholder(pil_style):
|
122 |
+
pil_style = None
|
123 |
+
|
124 |
+
first_faceid_url = []
|
125 |
+
if pil_faceid_1st:
|
126 |
+
first_faceid_url.append(get_image_url(request_id, pil_faceid_1st, suffix="_faceid_1st.png"))
|
127 |
+
if pil_supp_faceids_1st and len(pil_supp_faceids_1st) > 0:
|
128 |
+
for idx, pil_supp_faceid_1st in enumerate(pil_supp_faceids_1st):
|
129 |
+
pil_supp_faceid_1st = Image.open(pil_supp_faceid_1st)
|
130 |
+
first_faceid_url.append(get_image_url(request_id, pil_supp_faceid_1st, suffix=f"_faceid_1st_{idx}.png"))
|
131 |
+
|
132 |
+
second_faceid_url = []
|
133 |
+
if pil_faceid_2nd:
|
134 |
+
second_faceid_url.append(get_image_url(request_id, pil_faceid_2nd, suffix="_faceid_2nd.png"))
|
135 |
+
if pil_supp_faceids_2nd and len(pil_supp_faceids_2nd) > 0:
|
136 |
+
for idx, pil_supp_faceid_2nd in enumerate(pil_supp_faceids_2nd):
|
137 |
+
pil_supp_faceid_2nd = Image.open(pil_supp_faceid_2nd)
|
138 |
+
second_faceid_url.append(get_image_url(request_id, pil_supp_faceid_2nd, suffix=f"_faceid_2nd_{idx}.png"))
|
139 |
+
|
140 |
+
first_mix_face_url, first_mix_scale = [], []
|
141 |
+
if pil_mix_faceid_1_1st:
|
142 |
+
first_mix_face_url.append(
|
143 |
+
get_image_url(request_id, pil_mix_faceid_1_1st, suffix=f"_mix_faceid_1_1st_{mix_scale_1_1st:.2f}.png")
|
144 |
+
)
|
145 |
+
first_mix_scale.append(mix_scale_1_1st)
|
146 |
+
|
147 |
+
if pil_mix_faceid_2_1st is not None:
|
148 |
+
first_mix_face_url.append(
|
149 |
+
get_image_url(request_id, pil_mix_faceid_2_1st, suffix=f"_mix_faceid_2_1st_{mix_scale_2_1st:.2f}.png")
|
150 |
+
)
|
151 |
+
first_mix_scale.append(mix_scale_2_1st)
|
152 |
+
|
153 |
+
second_mix_face_url, second_mix_scale = [], []
|
154 |
+
if pil_mix_faceid_1_2nd:
|
155 |
+
second_mix_face_url.append(
|
156 |
+
get_image_url(request_id, pil_mix_faceid_1_2nd, suffix=f"_mix_faceid_1_2nd_{mix_scale_1_2nd:.2f}.png")
|
157 |
+
)
|
158 |
+
second_mix_scale.append(mix_scale_1_2nd)
|
159 |
+
|
160 |
+
if pil_mix_faceid_2_2nd is not None:
|
161 |
+
second_mix_face_url.append(
|
162 |
+
get_image_url(request_id, pil_mix_faceid_2_2nd, suffix=f"_mix_faceid_2_2nd_{mix_scale_2_2nd:.2f}.png")
|
163 |
+
)
|
164 |
+
second_mix_scale.append(mix_scale_2_2nd)
|
165 |
+
|
166 |
+
# The outer parenthesis represent the ID, and the inner parenthesis represent the mix face image for each ID.
|
167 |
+
faceid_image_url, mix_faceid_image_url, mix_faceid_scale = [], [], []
|
168 |
+
if len(first_faceid_url) > 0:
|
169 |
+
faceid_image_url.append(first_faceid_url)
|
170 |
+
mix_faceid_image_url.append(first_mix_face_url)
|
171 |
+
mix_faceid_scale.append(first_mix_scale)
|
172 |
+
if len(second_faceid_url) > 0:
|
173 |
+
faceid_image_url.append(second_faceid_url)
|
174 |
+
mix_faceid_image_url.append(second_mix_face_url)
|
175 |
+
mix_faceid_scale.append(second_mix_scale)
|
176 |
+
|
177 |
+
if pil_style is None:
|
178 |
+
style_image_url = ""
|
179 |
+
else:
|
180 |
+
style_image_url = get_image_url(request_id, pil_style, suffix="_style.png")
|
181 |
+
|
182 |
+
data = dict(
|
183 |
+
input=dict(
|
184 |
+
faceid_image_url=faceid_image_url,
|
185 |
+
mix_faceid_image_url=mix_faceid_image_url,
|
186 |
+
mix_faceid_scale=mix_faceid_scale,
|
187 |
+
style_image_url=style_image_url,
|
188 |
+
prompt=prompt,
|
189 |
+
negative_prompt=negative_prompt
|
190 |
+
),
|
191 |
+
parameters=dict(
|
192 |
+
image_resolution=image_resolution,
|
193 |
+
faceid_scale=faceid_scale * 10.,
|
194 |
+
face_structure_scale=face_structure_scale * 10.,
|
195 |
+
style_scale=style_scale * 10.,
|
196 |
+
use_sr=int(use_sr),
|
197 |
+
seed=seed,
|
198 |
+
mode="text-to-image",
|
199 |
+
)
|
200 |
+
)
|
201 |
+
|
202 |
+
res = call_generation(data)
|
203 |
+
|
204 |
+
return res
|
205 |
+
|
206 |
+
|
207 |
+
def image_to_single_id_generation_process(
|
208 |
+
pil_faceid=None, pil_supp_faceids=None,
|
209 |
+
pil_mix_faceid_1=None, mix_scale_1=0.0,
|
210 |
+
pil_mix_faceid_2=None, mix_scale_2=0.0,
|
211 |
+
faceid_scale=0.0, face_structure_scale=0.0,
|
212 |
+
pil_style=None, style_scale=1.0,
|
213 |
+
seed=-1, image_resolution="768x512", use_sr=True,
|
214 |
+
):
|
215 |
+
request_id = time.strftime('%Y%m%d-', time.localtime(time.time())) + str(uuid.uuid4())
|
216 |
+
|
217 |
+
if pil_faceid is None:
|
218 |
+
raise gr.Error("Please upload an ID image")
|
219 |
+
|
220 |
+
if pil_style is None:
|
221 |
+
raise gr.Error("Please upload a reference image")
|
222 |
+
|
223 |
+
faceid_url = []
|
224 |
+
if pil_faceid:
|
225 |
+
faceid_url.append(get_image_url(request_id, pil_faceid, suffix="_faceid.png"))
|
226 |
+
if pil_supp_faceids and len(pil_supp_faceids) > 0:
|
227 |
+
for idx, pil_supp_faceid in enumerate(pil_supp_faceids):
|
228 |
+
pil_supp_faceid = Image.open(pil_supp_faceid)
|
229 |
+
faceid_url.append(get_image_url(request_id, pil_supp_faceid, suffix=f"_supp_faceid_{idx}.png"))
|
230 |
+
|
231 |
+
mix_face_url = []
|
232 |
+
mix_scale = []
|
233 |
+
if pil_mix_faceid_1 is not None:
|
234 |
+
mix_face_url.append(
|
235 |
+
get_image_url(request_id, pil_mix_faceid_1, suffix=f"_mix_faceid_1_{mix_scale_1:.2f}.png")
|
236 |
+
)
|
237 |
+
mix_scale.append(mix_scale_1)
|
238 |
+
|
239 |
+
if pil_mix_faceid_2 is not None:
|
240 |
+
mix_face_url.append(
|
241 |
+
get_image_url(request_id, pil_mix_faceid_2, suffix=f"_mix_faceid_2_{mix_scale_2:.2f}.png")
|
242 |
+
)
|
243 |
+
mix_scale.append(mix_scale_2)
|
244 |
+
|
245 |
+
# The outer parenthesis represent the ID, and the inner parenthesis represent the mix face image for each ID.
|
246 |
+
faceid_image_url, mix_faceid_image_url, mix_faceid_scale = [], [], []
|
247 |
+
if len(faceid_url) > 0:
|
248 |
+
faceid_image_url.append(faceid_url)
|
249 |
+
mix_faceid_image_url.append(mix_face_url)
|
250 |
+
mix_faceid_scale.append(mix_scale)
|
251 |
+
|
252 |
+
if pil_style is None:
|
253 |
+
style_image_url = ""
|
254 |
+
else:
|
255 |
+
style_image_url = get_image_url(request_id, pil_style, suffix="_style.png")
|
256 |
+
|
257 |
+
data = dict(
|
258 |
+
input=dict(
|
259 |
+
faceid_image_url=faceid_image_url,
|
260 |
+
mix_faceid_image_url=mix_faceid_image_url,
|
261 |
+
mix_faceid_scale=mix_faceid_scale,
|
262 |
+
style_image_url=style_image_url,
|
263 |
+
),
|
264 |
+
parameters=dict(
|
265 |
+
image_resolution=image_resolution,
|
266 |
+
faceid_scale=faceid_scale * 10.,
|
267 |
+
face_structure_scale=face_structure_scale * 10.,
|
268 |
+
style_scale=style_scale * 10.,
|
269 |
+
use_sr=int(use_sr),
|
270 |
+
seed=seed,
|
271 |
+
mode="image-to-image",
|
272 |
+
)
|
273 |
+
)
|
274 |
+
|
275 |
+
res = call_generation(data)
|
276 |
+
|
277 |
+
return res
|
src/util.py
ADDED
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import concurrent.futures
|
2 |
+
import io
|
3 |
+
import os
|
4 |
+
|
5 |
+
import oss2
|
6 |
+
import requests
|
7 |
+
from PIL import Image
|
8 |
+
|
9 |
+
from .log import logger
|
10 |
+
|
11 |
+
# oss
|
12 |
+
access_key_id = os.getenv("ACCESS_KEY_ID")
|
13 |
+
access_key_secret = os.getenv("ACCESS_KEY_SECRET")
|
14 |
+
bucket_name = os.getenv("BUCKET_NAME")
|
15 |
+
endpoint = os.getenv("ENDPOINT")
|
16 |
+
|
17 |
+
bucket = oss2.Bucket(oss2.Auth(access_key_id, access_key_secret), endpoint, bucket_name)
|
18 |
+
oss_path = os.getenv("OSS_PATH")
|
19 |
+
|
20 |
+
|
21 |
+
def download_img_pil(index, img_url):
|
22 |
+
r = requests.get(img_url, stream=True)
|
23 |
+
if r.status_code == 200:
|
24 |
+
img = Image.open(io.BytesIO(r.content))
|
25 |
+
return (index, img)
|
26 |
+
else:
|
27 |
+
logger.error(f"Fail to download: {img_url}")
|
28 |
+
|
29 |
+
|
30 |
+
def download_images(img_urls, batch_size):
|
31 |
+
imgs_pil = [None] * batch_size
|
32 |
+
with concurrent.futures.ThreadPoolExecutor(max_workers=4) as executor:
|
33 |
+
to_do = []
|
34 |
+
for i, url in enumerate(img_urls):
|
35 |
+
future = executor.submit(download_img_pil, i, url)
|
36 |
+
to_do.append(future)
|
37 |
+
|
38 |
+
for future in concurrent.futures.as_completed(to_do):
|
39 |
+
ret = future.result()
|
40 |
+
index, img_pil = ret
|
41 |
+
imgs_pil[index] = img_pil
|
42 |
+
|
43 |
+
return imgs_pil
|
44 |
+
|
45 |
+
|
46 |
+
def upload_np_2_oss(input_image, name="cache.png"):
|
47 |
+
assert name.lower().endswith((".png", ".jpg")), name
|
48 |
+
imgByteArr = io.BytesIO()
|
49 |
+
if name.lower().endswith(".png"):
|
50 |
+
Image.fromarray(input_image).save(imgByteArr, format="PNG")
|
51 |
+
else:
|
52 |
+
Image.fromarray(input_image).save(imgByteArr, format="JPEG", quality=95)
|
53 |
+
imgByteArr = imgByteArr.getvalue()
|
54 |
+
|
55 |
+
bucket.put_object(oss_path + "/" + name, imgByteArr)
|
56 |
+
ret = bucket.sign_url('GET', oss_path + "/" + name, 60 * 60 * 24)
|
57 |
+
del imgByteArr
|
58 |
+
return ret
|