Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from PIL import Image
|
3 |
+
import torch
|
4 |
+
import skvideo.io
|
5 |
+
from diffusers import I2VGenXLPipeline
|
6 |
+
from diffusers.utils import export_to_video, load_image
|
7 |
+
import numpy as np
|
8 |
+
import imageio
|
9 |
+
from moviepy.editor import ImageSequenceClip
|
10 |
+
from transformers import MusicgenForConditionalGeneration, AutoProcessor
|
11 |
+
from scipy.io import wavfile
|
12 |
+
import ffmpeg
|
13 |
+
|
14 |
+
def generate_video(image, prompt, negative_prompt, video_length):
|
15 |
+
generator = torch.manual_seed(8888)
|
16 |
+
device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
|
17 |
+
print(f"Using device: {device}")
|
18 |
+
pipeline = I2VGenXLPipeline.from_pretrained("ali-vilab/i2vgen-xl", torch_dtype=torch.float32)
|
19 |
+
pipeline.to(device)
|
20 |
+
frames = pipeline(
|
21 |
+
prompt=prompt,
|
22 |
+
image=image,
|
23 |
+
num_inference_steps=2,
|
24 |
+
negative_prompt=negative_prompt,
|
25 |
+
guidance_scale=9.0,
|
26 |
+
generator=generator,
|
27 |
+
num_frames=video_length*20
|
28 |
+
).frames[0]
|
29 |
+
return frames
|
30 |
+
|
31 |
+
def export_frames_to_video(frames, output_file):
|
32 |
+
frames_np = [np.array(frame) for frame in frames]
|
33 |
+
clip = ImageSequenceClip(frames_np, fps=30)
|
34 |
+
clip.write_videofile(output_file, codec='libx264', audio=False)
|
35 |
+
|
36 |
+
def generate_music(prompt, unconditional=False):
|
37 |
+
model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
|
38 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
39 |
+
model.to(device)
|
40 |
+
if unconditional:
|
41 |
+
unconditional_inputs = model.get_unconditional_inputs(num_samples=1)
|
42 |
+
audio_values = model.generate(**unconditional_inputs, do_sample=True, max_new_tokens=256)
|
43 |
+
else:
|
44 |
+
processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
|
45 |
+
inputs = processor(
|
46 |
+
text=prompt,
|
47 |
+
padding=True,
|
48 |
+
return_tensors="pt",
|
49 |
+
)
|
50 |
+
audio_values = model.generate(**inputs.to(device), do_sample=True, guidance_scale=3, max_new_tokens=256)
|
51 |
+
sampling_rate = model.config.audio_encoder.sampling_rate
|
52 |
+
return audio_values[0].cpu().numpy(), sampling_rate
|
53 |
+
|
54 |
+
def combine_audio_video(audio_file, video_file, output_file):
|
55 |
+
audio = ffmpeg.input(audio_file)
|
56 |
+
video = ffmpeg.input(video_file)
|
57 |
+
output = ffmpeg.output(video, audio, output_file, vcodec='copy', acodec='aac')
|
58 |
+
ffmpeg.run(output)
|
59 |
+
|
60 |
+
st.title("AI-Powered Video and Music Generation")
|
61 |
+
|
62 |
+
st.sidebar.title("Options")
|
63 |
+
|
64 |
+
st.sidebar.subheader("Video Generation")
|
65 |
+
image = st.sidebar.file_uploader("Upload an image", type=["jpg", "png"])
|
66 |
+
prompt = st.sidebar.text_input("Enter the prompt")
|
67 |
+
negative_prompt = st.sidebar.text_input("Enter the negative prompt")
|
68 |
+
video_length = st.sidebar.number_input("Enter the video length (seconds)", min_value=1, value=10)
|
69 |
+
|
70 |
+
st.sidebar.subheader("Music Generation")
|
71 |
+
music_prompt = st.sidebar.text_input("Enter the music prompt")
|
72 |
+
unconditional = st.sidebar.checkbox("Generate unconditional music")
|
73 |
+
|
74 |
+
if st.sidebar.button("Generate Video and Music"):
|
75 |
+
if image is not None:
|
76 |
+
image = Image.open(image)
|
77 |
+
frames = generate_video(image, prompt, negative_prompt, video_length)
|
78 |
+
export_frames_to_video(frames, "output_video.mp4")
|
79 |
+
st.video("output_video.mp4")
|
80 |
+
|
81 |
+
audio_values, sampling_rate = generate_music(music_prompt, unconditional)
|
82 |
+
wavfile.write("musicgen_out.wav", sampling_rate, audio_values)
|
83 |
+
st.audio("musicgen_out.wav")
|
84 |
+
|
85 |
+
combine_audio_video("musicgen_out.wav", "output_video.mp4", "combined_output.mp4")
|
86 |
+
st.video("combined_output.mp4")
|