Spaces:
Running
on
L4
Running
on
L4
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
import sys
|
4 |
+
import shutil
|
5 |
+
import uuid
|
6 |
+
import subprocess
|
7 |
+
from glob import glob
|
8 |
+
from huggingface_hub import snapshot_download
|
9 |
+
|
10 |
+
# Download models
|
11 |
+
os.makedirs("checkpoints", exist_ok=True)
|
12 |
+
|
13 |
+
snapshot_download(
|
14 |
+
repo_id = "chunyu-li/LatentSync",
|
15 |
+
local_dir = "./checkpoints"
|
16 |
+
)
|
17 |
+
|
18 |
+
import argparse
|
19 |
+
from omegaconf import OmegaConf
|
20 |
+
import torch
|
21 |
+
from diffusers import AutoencoderKL, DDIMScheduler
|
22 |
+
from latentsync.models.unet import UNet3DConditionModel
|
23 |
+
from latentsync.pipelines.lipsync_pipeline import LipsyncPipeline
|
24 |
+
from diffusers.utils.import_utils import is_xformers_available
|
25 |
+
from accelerate.utils import set_seed
|
26 |
+
from latentsync.whisper.audio2feature import Audio2Feature
|
27 |
+
|
28 |
+
|
29 |
+
def main(video_path, audio_path, progress=gr.Progress(track_tqdm=True)):
|
30 |
+
inference_ckpt_path = "checkpoints/latentsync_unet.pt"
|
31 |
+
unet_config_path = "configs/unet/second_stage.yaml"
|
32 |
+
config = OmegaConf.load(unet_config_path)
|
33 |
+
|
34 |
+
print(f"Input video path: {video_path}")
|
35 |
+
print(f"Input audio path: {audio_path}")
|
36 |
+
print(f"Loaded checkpoint path: {inference_ckpt_path}")
|
37 |
+
|
38 |
+
scheduler = DDIMScheduler.from_pretrained("configs")
|
39 |
+
|
40 |
+
if config.model.cross_attention_dim == 768:
|
41 |
+
whisper_model_path = "checkpoints/whisper/small.pt"
|
42 |
+
elif config.model.cross_attention_dim == 384:
|
43 |
+
whisper_model_path = "checkpoints/whisper/tiny.pt"
|
44 |
+
else:
|
45 |
+
raise NotImplementedError("cross_attention_dim must be 768 or 384")
|
46 |
+
|
47 |
+
audio_encoder = Audio2Feature(model_path=whisper_model_path, device="cuda", num_frames=config.data.num_frames)
|
48 |
+
|
49 |
+
vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16)
|
50 |
+
vae.config.scaling_factor = 0.18215
|
51 |
+
vae.config.shift_factor = 0
|
52 |
+
|
53 |
+
unet, _ = UNet3DConditionModel.from_pretrained(
|
54 |
+
OmegaConf.to_container(config.model),
|
55 |
+
inference_ckpt_path, # load checkpoint
|
56 |
+
device="cpu",
|
57 |
+
)
|
58 |
+
|
59 |
+
unet = unet.to(dtype=torch.float16)
|
60 |
+
|
61 |
+
# set xformers
|
62 |
+
if is_xformers_available():
|
63 |
+
unet.enable_xformers_memory_efficient_attention()
|
64 |
+
|
65 |
+
pipeline = LipsyncPipeline(
|
66 |
+
vae=vae,
|
67 |
+
audio_encoder=audio_encoder,
|
68 |
+
unet=unet,
|
69 |
+
scheduler=scheduler,
|
70 |
+
).to("cuda")
|
71 |
+
|
72 |
+
if seed != -1:
|
73 |
+
set_seed(seed)
|
74 |
+
else:
|
75 |
+
torch.seed()
|
76 |
+
|
77 |
+
print(f"Initial seed: {torch.initial_seed()}")
|
78 |
+
|
79 |
+
unique_id = str(uuid.uuid4())
|
80 |
+
video_out_path = f"video_out{unique_id}.mp4"
|
81 |
+
|
82 |
+
pipeline(
|
83 |
+
video_path=video_path,
|
84 |
+
audio_path=audio_path,
|
85 |
+
video_out_path=video_out_path,
|
86 |
+
video_mask_path=video_out_path.replace(".mp4", "_mask.mp4"),
|
87 |
+
num_frames=config.data.num_frames,
|
88 |
+
num_inference_steps=config.run.inference_steps,
|
89 |
+
guidance_scale=1.0,
|
90 |
+
weight_dtype=torch.float16,
|
91 |
+
width=config.data.resolution,
|
92 |
+
height=config.data.resolution,
|
93 |
+
)
|
94 |
+
|
95 |
+
return video_out_path
|
96 |
+
|
97 |
+
|
98 |
+
css="""
|
99 |
+
div#col-container{
|
100 |
+
margin: 0 auto;
|
101 |
+
max-width: 982px;
|
102 |
+
}
|
103 |
+
"""
|
104 |
+
with gr.Blocks(css=css) as demo:
|
105 |
+
with gr.Column(elem_id="col-container"):
|
106 |
+
gr.Markdown("# LatentSync: Audio Conditioned Latent Diffusion Models for Lip Sync")
|
107 |
+
gr.Markdown("LatentSync, an end-to-end lip sync framework based on audio conditioned latent diffusion models without any intermediate motion representation, diverging from previous diffusion-based lip sync methods based on pixel space diffusion or two-stage generation.")
|
108 |
+
gr.HTML("""
|
109 |
+
<div style="display:flex;column-gap:4px;">
|
110 |
+
<a href="https://github.com/bytedance/LatentSync">
|
111 |
+
<img src='https://img.shields.io/badge/GitHub-Repo-blue'>
|
112 |
+
</a>
|
113 |
+
<a href="https://arxiv.org/abs/2412.09262">
|
114 |
+
<img src='https://img.shields.io/badge/ArXiv-Paper-red'>
|
115 |
+
</a>
|
116 |
+
<a href="https://huggingface.co/spaces/fffiloni/LatentSync?duplicate=true">
|
117 |
+
<img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-sm.svg" alt="Duplicate this Space">
|
118 |
+
</a>
|
119 |
+
<a href="https://huggingface.co/fffiloni">
|
120 |
+
<img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/follow-me-on-HF-sm-dark.svg" alt="Follow me on HF">
|
121 |
+
</a>
|
122 |
+
</div>
|
123 |
+
""")
|
124 |
+
with gr.Row():
|
125 |
+
with gr.Column():
|
126 |
+
video_input = gr.Video(label="Video Control", format="mp4")
|
127 |
+
audio_input = gr.Video(label="Audio Inpit", type="filepath")
|
128 |
+
submit_btn = gr.Button("Submit")
|
129 |
+
with gr.Column():
|
130 |
+
video_result = gr.Video(label="Result")
|
131 |
+
|
132 |
+
gr.Examples(
|
133 |
+
examples = [
|
134 |
+
["assets/demo1_video.mp4", "assets/demo1_audio.wav"],
|
135 |
+
["assets/demo2_video.mp4", "assets/demo2_audio.wav"],
|
136 |
+
["assets/demo3_video.mp4", "assets/demo3_audio.wav"],
|
137 |
+
],
|
138 |
+
inputs = [video_input, audio_input]
|
139 |
+
)
|
140 |
+
|
141 |
+
submit_btn.click(
|
142 |
+
fn = main,
|
143 |
+
inputs = [video_input, audio_input],
|
144 |
+
outputs = [video_result]
|
145 |
+
)
|
146 |
+
|
147 |
+
demo.queue().launch(show_api=False, show_error=True)
|