Spaces:
Sleeping
Sleeping
it works
Browse files
app.py
CHANGED
@@ -9,8 +9,222 @@ from huggingface_hub import snapshot_download
|
|
9 |
os.makedirs("checkpoints", exist_ok=True)
|
10 |
snapshot_download("alex4727/InstantDrag", local_dir="./checkpoints")
|
11 |
|
12 |
-
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
-
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
os.makedirs("checkpoints", exist_ok=True)
|
10 |
snapshot_download("alex4727/InstantDrag", local_dir="./checkpoints")
|
11 |
|
12 |
+
from demo.demo_utils import (
|
13 |
+
process_img,
|
14 |
+
get_points,
|
15 |
+
undo_points_image,
|
16 |
+
clear_all,
|
17 |
+
InstantDragPipeline,
|
18 |
+
)
|
19 |
|
20 |
+
LENGTH = 480 # Length of the square area displaying/editing images
|
21 |
+
|
22 |
+
with gr.Blocks() as demo:
|
23 |
+
pipeline = InstantDragPipeline(seed=42, device="cuda", dtype=torch.float16)
|
24 |
+
|
25 |
+
with gr.Row():
|
26 |
+
gr.Markdown(
|
27 |
+
"""
|
28 |
+
# InstantDrag: Improving Interactivity in Drag-based Image Editing
|
29 |
+
"""
|
30 |
+
)
|
31 |
+
with gr.Tab(label="InstantDrag Demo"):
|
32 |
+
selected_points = gr.State([]) # Store points
|
33 |
+
original_image = gr.State(value=None) # Store original input image
|
34 |
+
with gr.Row():
|
35 |
+
# Upload & Preprocess Image Column
|
36 |
+
with gr.Column():
|
37 |
+
gr.Markdown(
|
38 |
+
"""<p style="text-align: center; font-size: 20px">Upload & Preprocess Image</p>"""
|
39 |
+
)
|
40 |
+
canvas = gr.ImageEditor(
|
41 |
+
height=LENGTH,
|
42 |
+
width=LENGTH,
|
43 |
+
type="numpy",
|
44 |
+
image_mode="RGB",
|
45 |
+
label="Preprocess Image",
|
46 |
+
show_label=True,
|
47 |
+
interactive=True,
|
48 |
+
)
|
49 |
+
with gr.Row():
|
50 |
+
save_results = gr.Checkbox(
|
51 |
+
value=False,
|
52 |
+
label="Save Results",
|
53 |
+
scale=1,
|
54 |
+
)
|
55 |
+
undo_button = gr.Button("Undo Clicked Points", scale=3)
|
56 |
+
|
57 |
+
# Click Points Column
|
58 |
+
with gr.Column():
|
59 |
+
gr.Markdown(
|
60 |
+
"""<p style="text-align: center; font-size: 20px">Click Points</p>"""
|
61 |
+
)
|
62 |
+
input_image = gr.Image(
|
63 |
+
type="numpy",
|
64 |
+
label="Click Points",
|
65 |
+
show_label=True,
|
66 |
+
height=LENGTH,
|
67 |
+
width=LENGTH,
|
68 |
+
interactive=False,
|
69 |
+
show_fullscreen_button=False,
|
70 |
+
)
|
71 |
+
with gr.Row():
|
72 |
+
run_button = gr.Button("Run")
|
73 |
+
|
74 |
+
# Editing Results Column
|
75 |
+
with gr.Column():
|
76 |
+
gr.Markdown(
|
77 |
+
"""<p style="text-align: center; font-size: 20px">Editing Results</p>"""
|
78 |
+
)
|
79 |
+
edited_image = gr.Image(
|
80 |
+
type="numpy",
|
81 |
+
label="Editing Results",
|
82 |
+
show_label=True,
|
83 |
+
height=LENGTH,
|
84 |
+
width=LENGTH,
|
85 |
+
interactive=False,
|
86 |
+
show_fullscreen_button=False,
|
87 |
+
)
|
88 |
+
with gr.Row():
|
89 |
+
clear_all_button = gr.Button("Clear All")
|
90 |
+
with gr.Tab("Configs - make sure to check README for details"):
|
91 |
+
with gr.Row():
|
92 |
+
with gr.Column():
|
93 |
+
with gr.Row():
|
94 |
+
flowgen_choices = sorted(
|
95 |
+
[model for model in os.listdir("checkpoints/") if "flowgen" in model]
|
96 |
+
)
|
97 |
+
flowgen_ckpt = gr.Dropdown(
|
98 |
+
value=flowgen_choices[0],
|
99 |
+
label="Select FlowGen to use",
|
100 |
+
choices=flowgen_choices,
|
101 |
+
info="config2 for most cases, config3 for more fine-grained dragging",
|
102 |
+
scale=2,
|
103 |
+
)
|
104 |
+
flowdiffusion_choices = sorted(
|
105 |
+
[model for model in os.listdir("checkpoints/") if "flowdiffusion" in model]
|
106 |
+
)
|
107 |
+
flowdiffusion_ckpt = gr.Dropdown(
|
108 |
+
value=flowdiffusion_choices[0],
|
109 |
+
label="Select FlowDiffusion to use",
|
110 |
+
choices=flowdiffusion_choices,
|
111 |
+
info="single model for all cases",
|
112 |
+
scale=1,
|
113 |
+
)
|
114 |
+
image_guidance = gr.Number(
|
115 |
+
value=1.5,
|
116 |
+
label="Image Guidance Scale",
|
117 |
+
precision=2,
|
118 |
+
step=0.1,
|
119 |
+
scale=1,
|
120 |
+
info="typically between 1.0-2.0.",
|
121 |
+
)
|
122 |
+
flow_guidance = gr.Number(
|
123 |
+
value=1.5,
|
124 |
+
label="Flow Guidance Scale",
|
125 |
+
precision=2,
|
126 |
+
step=0.1,
|
127 |
+
scale=1,
|
128 |
+
info="typically between 1.0-5.0",
|
129 |
+
)
|
130 |
+
num_steps = gr.Number(
|
131 |
+
value=20,
|
132 |
+
label="Inference Steps",
|
133 |
+
precision=0,
|
134 |
+
step=1,
|
135 |
+
scale=1,
|
136 |
+
info="typically between 20-50, 20 is usually enough",
|
137 |
+
)
|
138 |
+
flowgen_output_scale = gr.Number(
|
139 |
+
value=-1.0,
|
140 |
+
label="FlowGen Output Scale",
|
141 |
+
precision=1,
|
142 |
+
step=0.1,
|
143 |
+
scale=2,
|
144 |
+
info="-1.0, by default, forces flowgen's output to [-1, 1], could be adjusted to [0, ∞] for stronger/weaker effects",
|
145 |
+
)
|
146 |
+
gr.Markdown(
|
147 |
+
"""
|
148 |
+
<p style="text-align: center; font-size: 18px;">Examples</p>
|
149 |
+
"""
|
150 |
+
)
|
151 |
+
with gr.Row():
|
152 |
+
gr.Examples(
|
153 |
+
examples=[
|
154 |
+
"/home/user/app/InstDrag/demo/samples/airplane.jpg",
|
155 |
+
"/home/user/app/InstDrag/demo/samples/anime.jpg",
|
156 |
+
"/home/user/app/InstDrag/demo/samples/caligraphy.jpg",
|
157 |
+
"/home/user/app/InstDrag/demo/samples/crocodile.jpg",
|
158 |
+
"/home/user/app/InstDrag/demo/samples/elephant.jpg",
|
159 |
+
"/home/user/app/InstDrag/demo/samples/meteor.jpg",
|
160 |
+
"/home/user/app/InstDrag/demo/samples/monalisa.jpg",
|
161 |
+
"/home/user/app/InstDrag/demo/samples/portrait.jpg",
|
162 |
+
"/home/user/app/InstDrag/demo/samples/sketch.jpg",
|
163 |
+
"/home/user/app/InstDrag/demo/samples/surreal.jpg",
|
164 |
+
],
|
165 |
+
inputs=[canvas],
|
166 |
+
outputs=[original_image, selected_points, input_image],
|
167 |
+
fn=process_img,
|
168 |
+
cache_examples=False,
|
169 |
+
examples_per_page=10,
|
170 |
+
)
|
171 |
+
gr.Markdown(
|
172 |
+
"""
|
173 |
+
<p style="text-align: center; font-size: 9">[Important] Our base models are solely trained on real-world talking head (facial) videos, with a focus on achieving fine-grained facial editing. <br>
|
174 |
+
Their application to other types of scenes, without fine-tuning, should be considered more of an experimental byproduct and may not perform well in many cases (we currently support only square images).</p>
|
175 |
+
"""
|
176 |
+
)
|
177 |
+
# Event Handlers
|
178 |
+
canvas.change(
|
179 |
+
process_img,
|
180 |
+
[canvas],
|
181 |
+
[original_image, selected_points, input_image],
|
182 |
+
)
|
183 |
+
|
184 |
+
input_image.select(
|
185 |
+
get_points,
|
186 |
+
[input_image, selected_points],
|
187 |
+
[input_image],
|
188 |
+
)
|
189 |
+
|
190 |
+
undo_button.click(
|
191 |
+
undo_points_image,
|
192 |
+
[original_image],
|
193 |
+
[input_image, selected_points],
|
194 |
+
)
|
195 |
+
|
196 |
+
run_button.click(
|
197 |
+
pipeline.run,
|
198 |
+
[
|
199 |
+
original_image,
|
200 |
+
selected_points,
|
201 |
+
flowgen_ckpt,
|
202 |
+
flowdiffusion_ckpt,
|
203 |
+
image_guidance,
|
204 |
+
flow_guidance,
|
205 |
+
flowgen_output_scale,
|
206 |
+
num_steps,
|
207 |
+
save_results,
|
208 |
+
],
|
209 |
+
[edited_image],
|
210 |
+
)
|
211 |
+
|
212 |
+
clear_all_button.click(
|
213 |
+
clear_all,
|
214 |
+
[],
|
215 |
+
[
|
216 |
+
canvas,
|
217 |
+
input_image,
|
218 |
+
edited_image,
|
219 |
+
selected_points,
|
220 |
+
original_image,
|
221 |
+
],
|
222 |
+
)
|
223 |
+
|
224 |
+
demo.queue().launch(ssr_mode=False)
|
225 |
+
|
226 |
+
# def greet(name):
|
227 |
+
# return "Hello " + name + "!!"
|
228 |
+
|
229 |
+
# demo = gr.Interface(fn=greet, inputs="text", outputs="text")
|
230 |
+
# demo.launch()
|