Spaces:
Running
on
Zero
Running
on
Zero
adding slider edits
Browse files
app.py
CHANGED
@@ -8,6 +8,9 @@ import torch
|
|
8 |
from diffusers import DiffusionPipeline, UNet2DConditionModel, LCMScheduler
|
9 |
from huggingface_hub import hf_hub_download
|
10 |
from safetensors.torch import load_file
|
|
|
|
|
|
|
11 |
|
12 |
model_repo_id = "stabilityai/stable-diffusion-xl-base-1.0"
|
13 |
repo_name = "tianweiy/DMD2"
|
@@ -27,6 +30,26 @@ pipe = DiffusionPipeline.from_pretrained(model_repo_id, unet=unet, torch_dtype=t
|
|
27 |
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
|
28 |
|
29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
MAX_SEED = np.iinfo(np.int32).max
|
32 |
MAX_IMAGE_SIZE = 1024
|
@@ -42,25 +65,49 @@ def infer(
|
|
42 |
height,
|
43 |
guidance_scale,
|
44 |
num_inference_steps,
|
|
|
|
|
45 |
progress=gr.Progress(track_tqdm=True),
|
46 |
):
|
47 |
if randomize_seed:
|
48 |
seed = random.randint(0, MAX_SEED)
|
49 |
|
50 |
-
|
|
|
51 |
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
image = pipe(
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
-
|
|
|
64 |
|
65 |
|
66 |
examples = [
|
@@ -91,7 +138,32 @@ with gr.Blocks(css=css) as demo:
|
|
91 |
|
92 |
run_button = gr.Button("Run", scale=0, variant="primary")
|
93 |
|
94 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
|
96 |
with gr.Accordion("Advanced Settings", open=False):
|
97 |
negative_prompt = gr.Text(
|
@@ -158,8 +230,10 @@ with gr.Blocks(css=css) as demo:
|
|
158 |
height,
|
159 |
guidance_scale,
|
160 |
num_inference_steps,
|
|
|
|
|
161 |
],
|
162 |
-
outputs=[result, seed],
|
163 |
)
|
164 |
|
165 |
if __name__ == "__main__":
|
|
|
8 |
from diffusers import DiffusionPipeline, UNet2DConditionModel, LCMScheduler
|
9 |
from huggingface_hub import hf_hub_download
|
10 |
from safetensors.torch import load_file
|
11 |
+
import sys
|
12 |
+
sys.path.append('.')
|
13 |
+
from utils.lora import LoRANetwork, DEFAULT_TARGET_REPLACE, UNET_TARGET_REPLACE_MODULE_CONV
|
14 |
|
15 |
model_repo_id = "stabilityai/stable-diffusion-xl-base-1.0"
|
16 |
repo_name = "tianweiy/DMD2"
|
|
|
30 |
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
|
31 |
|
32 |
|
33 |
+
unet = pipe.unet
|
34 |
+
|
35 |
+
## Change these parameters based on how you trained your sliderspace sliders
|
36 |
+
train_method = 'xattn-strict'
|
37 |
+
rank = 1
|
38 |
+
alpha =1
|
39 |
+
networks = {}
|
40 |
+
modules = DEFAULT_TARGET_REPLACE
|
41 |
+
modules += UNET_TARGET_REPLACE_MODULE_CONV
|
42 |
+
for i in range(1):
|
43 |
+
networks[i] = LoRANetwork(
|
44 |
+
unet,
|
45 |
+
rank=int(rank),
|
46 |
+
multiplier=1.0,
|
47 |
+
alpha=int(alpha),
|
48 |
+
train_method=train_method,
|
49 |
+
fast_init=True,
|
50 |
+
).to(device, dtype=torch_dtype)
|
51 |
+
|
52 |
+
|
53 |
|
54 |
MAX_SEED = np.iinfo(np.int32).max
|
55 |
MAX_IMAGE_SIZE = 1024
|
|
|
65 |
height,
|
66 |
guidance_scale,
|
67 |
num_inference_steps,
|
68 |
+
sliderspace_path,
|
69 |
+
slider_scale,
|
70 |
progress=gr.Progress(track_tqdm=True),
|
71 |
):
|
72 |
if randomize_seed:
|
73 |
seed = random.randint(0, MAX_SEED)
|
74 |
|
75 |
+
for net in networks:
|
76 |
+
networks[net].load_state_dict(torch.load(sliderspace_path))
|
77 |
|
78 |
+
for net in networks:
|
79 |
+
networks[net].set_lora_slider(slider_scale)
|
80 |
+
|
81 |
+
with networks[0]:
|
82 |
+
pass
|
83 |
+
|
84 |
+
# original image
|
85 |
+
generator = torch.Generator().manual_seed(seed)
|
86 |
image = pipe(
|
87 |
+
prompt=prompt,
|
88 |
+
negative_prompt=negative_prompt,
|
89 |
+
guidance_scale=guidance_scale,
|
90 |
+
num_inference_steps=num_inference_steps,
|
91 |
+
width=width,
|
92 |
+
height=height,
|
93 |
+
generator=generator,
|
94 |
+
).images[0]
|
95 |
+
|
96 |
+
# edited image
|
97 |
+
generator = torch.Generator().manual_seed(seed)
|
98 |
+
with networks[0]:
|
99 |
+
slider_image = pipe(
|
100 |
+
prompt=prompt,
|
101 |
+
negative_prompt=negative_prompt,
|
102 |
+
guidance_scale=guidance_scale,
|
103 |
+
num_inference_steps=num_inference_steps,
|
104 |
+
width=width,
|
105 |
+
height=height,
|
106 |
+
generator=generator,
|
107 |
+
).images[0]
|
108 |
|
109 |
+
|
110 |
+
return image, slider_image, seed
|
111 |
|
112 |
|
113 |
examples = [
|
|
|
138 |
|
139 |
run_button = gr.Button("Run", scale=0, variant="primary")
|
140 |
|
141 |
+
|
142 |
+
# New dropdowns side by side
|
143 |
+
with gr.Row():
|
144 |
+
slider_space = gr.Dropdown(
|
145 |
+
choices=["spaceship", "car", "person"],
|
146 |
+
label="SliderSpace",
|
147 |
+
value="spaceship"
|
148 |
+
)
|
149 |
+
discovered_directions = gr.Dropdown(
|
150 |
+
choices=[f"Slider {i}" for i in range(1, 11)],
|
151 |
+
label="Discovered Directions",
|
152 |
+
value="Slider 1"
|
153 |
+
)
|
154 |
+
|
155 |
+
slider_scale = gr.Slider(
|
156 |
+
label="Slider Scale",
|
157 |
+
minimum=-2,
|
158 |
+
maximum=2,
|
159 |
+
step=0.1,
|
160 |
+
value=1,
|
161 |
+
)
|
162 |
+
|
163 |
+
with gr.Row():
|
164 |
+
result = gr.Image(label="Original Image", show_label=True)
|
165 |
+
slider_result = gr.Image(label="Discovered Edit Direction", show_label=True)
|
166 |
+
|
167 |
|
168 |
with gr.Accordion("Advanced Settings", open=False):
|
169 |
negative_prompt = gr.Text(
|
|
|
230 |
height,
|
231 |
guidance_scale,
|
232 |
num_inference_steps,
|
233 |
+
f"sliderspace_weights/{slider_space}/{discovered_directions}",
|
234 |
+
slider_scale
|
235 |
],
|
236 |
+
outputs=[result, slider_result, seed],
|
237 |
)
|
238 |
|
239 |
if __name__ == "__main__":
|