Update app.py
Browse files
app.py
CHANGED
|
@@ -69,7 +69,7 @@ def get_modelscope_pipeline(
|
|
| 69 |
lora.merge_and_unload()
|
| 70 |
pipe.unet = lora
|
| 71 |
|
| 72 |
-
pipe = pipe.to(device)
|
| 73 |
|
| 74 |
return pipe
|
| 75 |
|
|
@@ -136,22 +136,48 @@ def get_animatediff_pipeline(
|
|
| 136 |
lora.merge_and_unload()
|
| 137 |
pipe.unet = lora
|
| 138 |
|
| 139 |
-
pipe = pipe.to(device)
|
| 140 |
return pipe
|
| 141 |
|
| 142 |
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
# }
|
| 149 |
-
cache_pipeline = {
|
| 150 |
-
"base_model": None,
|
| 151 |
-
"variant": None,
|
| 152 |
-
"pipeline": None,
|
| 153 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 154 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 155 |
|
| 156 |
@spaces.GPU(duration=120)
|
| 157 |
def infer(
|
|
@@ -180,45 +206,46 @@ def infer(
|
|
| 180 |
# )
|
| 181 |
# else:
|
| 182 |
# raise ValueError(f"Unknown base_model {base_model}")
|
| 183 |
-
if (
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
):
|
| 187 |
-
|
| 188 |
-
else:
|
| 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 |
if randomize_seed:
|
| 217 |
seed = random.randint(0, MAX_SEED)
|
| 218 |
|
| 219 |
generator = torch.Generator("cpu").manual_seed(seed)
|
| 220 |
|
| 221 |
-
output =
|
| 222 |
prompt=prompt,
|
| 223 |
num_frames=16,
|
| 224 |
guidance_scale=1.0,
|
|
@@ -238,7 +265,8 @@ def infer(
|
|
| 238 |
fps=7,
|
| 239 |
)
|
| 240 |
print(f"Saved to {save_path}")
|
| 241 |
-
|
|
|
|
| 242 |
|
| 243 |
|
| 244 |
examples = [
|
|
@@ -402,7 +430,7 @@ with gr.Blocks(css=css) as demo:
|
|
| 402 |
inputs=[base_model, variant_dropdown, prompt, num_inference_steps],
|
| 403 |
cache_examples=True,
|
| 404 |
fn=infer,
|
| 405 |
-
outputs=[result],
|
| 406 |
)
|
| 407 |
|
| 408 |
run_button.click(
|
|
@@ -415,7 +443,7 @@ with gr.Blocks(css=css) as demo:
|
|
| 415 |
seed,
|
| 416 |
randomize_seed,
|
| 417 |
],
|
| 418 |
-
outputs=[result],
|
| 419 |
)
|
| 420 |
|
| 421 |
demo.queue().launch()
|
|
|
|
| 69 |
lora.merge_and_unload()
|
| 70 |
pipe.unet = lora
|
| 71 |
|
| 72 |
+
# pipe = pipe.to(device)
|
| 73 |
|
| 74 |
return pipe
|
| 75 |
|
|
|
|
| 136 |
lora.merge_and_unload()
|
| 137 |
pipe.unet = lora
|
| 138 |
|
| 139 |
+
# pipe = pipe.to(device)
|
| 140 |
return pipe
|
| 141 |
|
| 142 |
|
| 143 |
+
pipe_dict = {
|
| 144 |
+
"ModelScope T2V": {"WebVid": None, "LAION-aes": None, "Anime": None, "Realistic": None, "3D Cartoon": None},
|
| 145 |
+
"AnimateDiff (SD1.5)": {"WebVid": None, "LAION-aes": None},
|
| 146 |
+
"AnimateDiff (RealisticVision)": {"WebVid": None, "LAION-aes": None},
|
| 147 |
+
"AnimateDiff (epiCRealism)": {"WebVid": None, "LAION-aes": None},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 148 |
}
|
| 149 |
+
# cache_pipeline = {
|
| 150 |
+
# "base_model": None,
|
| 151 |
+
# "variant": None,
|
| 152 |
+
# "pipeline": None,
|
| 153 |
+
# }
|
| 154 |
|
| 155 |
+
def init_pipelines():
|
| 156 |
+
for base_model in variants.keys():
|
| 157 |
+
for variant in variants[base_model]:
|
| 158 |
+
if pipe_dict[base_model][variant] is None:
|
| 159 |
+
if base_model == "ModelScope T2V":
|
| 160 |
+
pipe_dict[base_model][variant] = get_modelscope_pipeline(mcm_variant=variant)
|
| 161 |
+
elif base_model == "AnimateDiff (SD1.5)":
|
| 162 |
+
pipe_dict[base_model][variant] = get_animatediff_pipeline(
|
| 163 |
+
real_variant=None,
|
| 164 |
+
motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 165 |
+
mcm_variant=variant,
|
| 166 |
+
)
|
| 167 |
+
elif base_model == "AnimateDiff (RealisticVision)":
|
| 168 |
+
pipe_dict[base_model][variant] = get_animatediff_pipeline(
|
| 169 |
+
real_variant="realvision",
|
| 170 |
+
motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 171 |
+
mcm_variant=variant,
|
| 172 |
+
)
|
| 173 |
+
elif base_model == "AnimateDiff (epiCRealism)":
|
| 174 |
+
pipe_dict[base_model][variant] = get_animatediff_pipeline(
|
| 175 |
+
real_variant="epicrealism",
|
| 176 |
+
motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 177 |
+
mcm_variant=variant,
|
| 178 |
+
)
|
| 179 |
+
else:
|
| 180 |
+
raise ValueError(f"Unknown base_model {base_model}")
|
| 181 |
|
| 182 |
@spaces.GPU(duration=120)
|
| 183 |
def infer(
|
|
|
|
| 206 |
# )
|
| 207 |
# else:
|
| 208 |
# raise ValueError(f"Unknown base_model {base_model}")
|
| 209 |
+
# if (
|
| 210 |
+
# cache_pipeline["base_model"] == base_model
|
| 211 |
+
# and cache_pipeline["variant"] == variant
|
| 212 |
+
# ):
|
| 213 |
+
# pass
|
| 214 |
+
# else:
|
| 215 |
+
# if base_model == "ModelScope T2V":
|
| 216 |
+
# pipeline = get_modelscope_pipeline(mcm_variant=variant)
|
| 217 |
+
# elif base_model == "AnimateDiff (SD1.5)":
|
| 218 |
+
# pipeline = get_animatediff_pipeline(
|
| 219 |
+
# real_variant=None,
|
| 220 |
+
# motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 221 |
+
# mcm_variant=variant,
|
| 222 |
+
# )
|
| 223 |
+
# elif base_model == "AnimateDiff (RealisticVision)":
|
| 224 |
+
# pipeline = get_animatediff_pipeline(
|
| 225 |
+
# real_variant="realvision",
|
| 226 |
+
# motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 227 |
+
# mcm_variant=variant,
|
| 228 |
+
# )
|
| 229 |
+
# elif base_model == "AnimateDiff (epiCRealism)":
|
| 230 |
+
# pipeline = get_animatediff_pipeline(
|
| 231 |
+
# real_variant="epicrealism",
|
| 232 |
+
# motion_module_path="guoyww/animatediff-motion-adapter-v1-5-2",
|
| 233 |
+
# mcm_variant=variant,
|
| 234 |
+
# )
|
| 235 |
+
# else:
|
| 236 |
+
# raise ValueError(f"Unknown base_model {base_model}")
|
| 237 |
|
| 238 |
+
# cache_pipeline["base_model"] = base_model
|
| 239 |
+
# cache_pipeline["variant"] = variant
|
| 240 |
+
# cache_pipeline["pipeline"] = pipeline
|
| 241 |
|
| 242 |
+
pipe_dict[base_model][variant] = pipe_dict[base_model][variant].to(device)
|
| 243 |
if randomize_seed:
|
| 244 |
seed = random.randint(0, MAX_SEED)
|
| 245 |
|
| 246 |
generator = torch.Generator("cpu").manual_seed(seed)
|
| 247 |
|
| 248 |
+
output = pipe_dict[base_model][variant](
|
| 249 |
prompt=prompt,
|
| 250 |
num_frames=16,
|
| 251 |
guidance_scale=1.0,
|
|
|
|
| 265 |
fps=7,
|
| 266 |
)
|
| 267 |
print(f"Saved to {save_path}")
|
| 268 |
+
pipe_dict[base_model][variant] = pipe_dict[base_model][variant].to("cpu")
|
| 269 |
+
return save_path, seed
|
| 270 |
|
| 271 |
|
| 272 |
examples = [
|
|
|
|
| 430 |
inputs=[base_model, variant_dropdown, prompt, num_inference_steps],
|
| 431 |
cache_examples=True,
|
| 432 |
fn=infer,
|
| 433 |
+
outputs=[result, seed],
|
| 434 |
)
|
| 435 |
|
| 436 |
run_button.click(
|
|
|
|
| 443 |
seed,
|
| 444 |
randomize_seed,
|
| 445 |
],
|
| 446 |
+
outputs=[result, seed],
|
| 447 |
)
|
| 448 |
|
| 449 |
demo.queue().launch()
|