Spaces:
Runtime error
Runtime error
Update
Browse files
app.py
CHANGED
|
@@ -11,53 +11,66 @@ from safetensors.torch import load_file
|
|
| 11 |
from PIL import Image
|
| 12 |
|
| 13 |
# Constants
|
| 14 |
-
|
| 15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
# Ensure model and scheduler are initialized in GPU-enabled function
|
| 18 |
-
if torch.cuda.is_available():
|
| 19 |
-
device = "cuda"
|
| 20 |
-
dtype = torch.float16
|
| 21 |
-
adapter = MotionAdapter().to(device, dtype)
|
| 22 |
-
pipe = AnimateDiffPipeline.from_pretrained(base, motion_adapter=adapter, torch_dtype=dtype).to(device)
|
| 23 |
-
pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing", beta_schedule="linear")
|
| 24 |
-
else:
|
| 25 |
raise NotImplementedError("No GPU detected!")
|
| 26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
# Function
|
| 28 |
@spaces.GPU(enable_queue=True)
|
| 29 |
-
def generate_image(prompt, step):
|
| 30 |
global loaded
|
| 31 |
print(prompt, step)
|
| 32 |
|
| 33 |
-
if
|
| 34 |
repo = "ByteDance/AnimateDiff-Lightning"
|
| 35 |
ckpt = f"animatediff_lightning_{step}step_diffusers.safetensors"
|
| 36 |
pipe.unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device=device), strict=False)
|
| 37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
output = pipe(prompt=prompt, guidance_scale=1.0, num_inference_steps=step)
|
| 40 |
-
|
| 41 |
name = str(uuid.uuid4()).replace("-", "")
|
| 42 |
path = f"/tmp/{name}.mp4"
|
| 43 |
export_to_video(output.frames[0], path, fps=10)
|
| 44 |
return path
|
| 45 |
|
| 46 |
|
| 47 |
-
|
| 48 |
# Gradio Interface
|
| 49 |
-
|
| 50 |
with gr.Blocks(css="style.css") as demo:
|
| 51 |
gr.HTML("<h1><center>AnimateDiff-Lightning ⚡</center></h1>")
|
| 52 |
gr.HTML("<p><center>Lightning-fast text-to-video generation</center></p><p><center><a href='https://huggingface.co/ByteDance/AnimateDiff-Lightning'>https://huggingface.co/ByteDance/AnimateDiff-Lightning</a></center></p>")
|
| 53 |
with gr.Group():
|
| 54 |
with gr.Row():
|
| 55 |
prompt = gr.Textbox(
|
| 56 |
-
label='
|
| 57 |
scale=8
|
| 58 |
)
|
| 59 |
-
|
| 60 |
-
label='
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
choices=[
|
| 62 |
('1-Step', 1),
|
| 63 |
('2-Step', 2),
|
|
@@ -77,12 +90,12 @@ with gr.Blocks(css="style.css") as demo:
|
|
| 77 |
|
| 78 |
prompt.submit(
|
| 79 |
fn=generate_image,
|
| 80 |
-
inputs=[prompt,
|
| 81 |
outputs=video,
|
| 82 |
)
|
| 83 |
submit.click(
|
| 84 |
fn=generate_image,
|
| 85 |
-
inputs=[prompt,
|
| 86 |
outputs=video,
|
| 87 |
)
|
| 88 |
|
|
|
|
| 11 |
from PIL import Image
|
| 12 |
|
| 13 |
# Constants
|
| 14 |
+
bases = {
|
| 15 |
+
"ToonYou": "frankjoshua/toonyou_beta6",
|
| 16 |
+
"epiCRealism": "emilianJR/epiCRealism"
|
| 17 |
+
}
|
| 18 |
+
step_loaded = None
|
| 19 |
+
base_loaded = "ToonYou"
|
| 20 |
|
| 21 |
# Ensure model and scheduler are initialized in GPU-enabled function
|
| 22 |
+
if not torch.cuda.is_available():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
raise NotImplementedError("No GPU detected!")
|
| 24 |
|
| 25 |
+
device = "cuda"
|
| 26 |
+
dtype = torch.float16
|
| 27 |
+
pipe = AnimateDiffPipeline.from_pretrained(bases[base_loaded], torch_dtype=dtype).to(device)
|
| 28 |
+
pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing", beta_schedule="linear")
|
| 29 |
+
|
| 30 |
# Function
|
| 31 |
@spaces.GPU(enable_queue=True)
|
| 32 |
+
def generate_image(prompt, base, step):
|
| 33 |
global loaded
|
| 34 |
print(prompt, step)
|
| 35 |
|
| 36 |
+
if step_loaded != step:
|
| 37 |
repo = "ByteDance/AnimateDiff-Lightning"
|
| 38 |
ckpt = f"animatediff_lightning_{step}step_diffusers.safetensors"
|
| 39 |
pipe.unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device=device), strict=False)
|
| 40 |
+
step_loaded = step
|
| 41 |
+
|
| 42 |
+
if base_loaded != base:
|
| 43 |
+
pipe.unet.load_state_dict(torch.load(hf_hub_download(bases[base], "unet/diffusion_pytorch_model.bin"), map_location=device), strict=False)
|
| 44 |
+
base_loaded = base
|
| 45 |
|
| 46 |
output = pipe(prompt=prompt, guidance_scale=1.0, num_inference_steps=step)
|
|
|
|
| 47 |
name = str(uuid.uuid4()).replace("-", "")
|
| 48 |
path = f"/tmp/{name}.mp4"
|
| 49 |
export_to_video(output.frames[0], path, fps=10)
|
| 50 |
return path
|
| 51 |
|
| 52 |
|
|
|
|
| 53 |
# Gradio Interface
|
|
|
|
| 54 |
with gr.Blocks(css="style.css") as demo:
|
| 55 |
gr.HTML("<h1><center>AnimateDiff-Lightning ⚡</center></h1>")
|
| 56 |
gr.HTML("<p><center>Lightning-fast text-to-video generation</center></p><p><center><a href='https://huggingface.co/ByteDance/AnimateDiff-Lightning'>https://huggingface.co/ByteDance/AnimateDiff-Lightning</a></center></p>")
|
| 57 |
with gr.Group():
|
| 58 |
with gr.Row():
|
| 59 |
prompt = gr.Textbox(
|
| 60 |
+
label='Prompt (English)',
|
| 61 |
scale=8
|
| 62 |
)
|
| 63 |
+
select_base = gr.Dropdown(
|
| 64 |
+
label='Base model',
|
| 65 |
+
choices=[
|
| 66 |
+
"ToonYou",
|
| 67 |
+
"epiCRealism",
|
| 68 |
+
],
|
| 69 |
+
value=base_loaded,
|
| 70 |
+
interactive=True
|
| 71 |
+
)
|
| 72 |
+
select_step = gr.Dropdown(
|
| 73 |
+
label='Inference steps',
|
| 74 |
choices=[
|
| 75 |
('1-Step', 1),
|
| 76 |
('2-Step', 2),
|
|
|
|
| 90 |
|
| 91 |
prompt.submit(
|
| 92 |
fn=generate_image,
|
| 93 |
+
inputs=[prompt, select_base, select_step],
|
| 94 |
outputs=video,
|
| 95 |
)
|
| 96 |
submit.click(
|
| 97 |
fn=generate_image,
|
| 98 |
+
inputs=[prompt, select_base, select_step],
|
| 99 |
outputs=video,
|
| 100 |
)
|
| 101 |
|