Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -6,6 +6,9 @@ from einops import rearrange
|
|
6 |
import gradio as gr
|
7 |
from stable_audio_tools import get_pretrained_model
|
8 |
from stable_audio_tools.inference.generation import generate_diffusion_cond
|
|
|
|
|
|
|
9 |
|
10 |
# Authenticate
|
11 |
token = os.getenv("HUGGINGFACE_TOKEN")
|
@@ -13,19 +16,25 @@ if not token:
|
|
13 |
raise RuntimeError("HUGGINGFACE_TOKEN not set")
|
14 |
login(token=token, add_to_git_credential=False)
|
15 |
|
16 |
-
# Load model
|
17 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
18 |
-
|
19 |
-
|
20 |
-
sample_rate =
|
21 |
-
sample_size =
|
22 |
|
23 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
def generate_audio(prompt):
|
25 |
conditioning = [{"prompt": prompt, "seconds_total": 11}]
|
26 |
with torch.no_grad():
|
27 |
output = generate_diffusion_cond(
|
28 |
-
|
29 |
steps=8,
|
30 |
conditioning=conditioning,
|
31 |
sample_size=sample_size,
|
@@ -37,24 +46,45 @@ def generate_audio(prompt):
|
|
37 |
torchaudio.save(path, output, sample_rate)
|
38 |
return path
|
39 |
|
40 |
-
#
|
41 |
-
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
inputs=gr.Textbox(
|
44 |
label="π€ Prompt your sonic art here",
|
45 |
placeholder="e.g. 'drunk driving with mario and yung lean'"
|
46 |
),
|
47 |
-
outputs=gr.
|
48 |
-
|
49 |
-
label="π§ Generated Audio"
|
50 |
),
|
51 |
title='π Hot Prompts in Your Area: "My Husband Is Dead"',
|
52 |
-
description="Enter a fun sound idea for music art.",
|
53 |
examples=[
|
54 |
"ghosts peeing in a server room",
|
55 |
"tech startup boss villain entrance music",
|
56 |
"AI doing acid in a technofeudalist dystopia"
|
57 |
],
|
58 |
css="style.css"
|
59 |
-
)
|
60 |
|
|
|
|
6 |
import gradio as gr
|
7 |
from stable_audio_tools import get_pretrained_model
|
8 |
from stable_audio_tools.inference.generation import generate_diffusion_cond
|
9 |
+
from diffusers import DiffusionPipeline
|
10 |
+
from PIL import Image
|
11 |
+
from moviepy.editor import AudioFileClip, ImageClip
|
12 |
|
13 |
# Authenticate
|
14 |
token = os.getenv("HUGGINGFACE_TOKEN")
|
|
|
16 |
raise RuntimeError("HUGGINGFACE_TOKEN not set")
|
17 |
login(token=token, add_to_git_credential=False)
|
18 |
|
19 |
+
# Load audio model
|
20 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
21 |
+
audio_model, audio_config = get_pretrained_model("stabilityai/stable-audio-open-small")
|
22 |
+
audio_model = audio_model.to(device)
|
23 |
+
sample_rate = audio_config["sample_rate"]
|
24 |
+
sample_size = audio_config["sample_size"]
|
25 |
|
26 |
+
# Load image model (Kandinsky)
|
27 |
+
image_pipe = DiffusionPipeline.from_pretrained(
|
28 |
+
"kandinsky-community/kandinsky-3",
|
29 |
+
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
|
30 |
+
).to(device)
|
31 |
+
|
32 |
+
# Generate audio
|
33 |
def generate_audio(prompt):
|
34 |
conditioning = [{"prompt": prompt, "seconds_total": 11}]
|
35 |
with torch.no_grad():
|
36 |
output = generate_diffusion_cond(
|
37 |
+
audio_model,
|
38 |
steps=8,
|
39 |
conditioning=conditioning,
|
40 |
sample_size=sample_size,
|
|
|
46 |
torchaudio.save(path, output, sample_rate)
|
47 |
return path
|
48 |
|
49 |
+
# Generate image
|
50 |
+
def generate_image(prompt):
|
51 |
+
image = image_pipe(prompt=prompt, height=500, width=500).images[0]
|
52 |
+
image_path = "output.png"
|
53 |
+
image.save(image_path)
|
54 |
+
return image_path
|
55 |
+
|
56 |
+
# Combine audio + image into mp4
|
57 |
+
def combine_to_video(image_path, audio_path, output_path="output.mp4"):
|
58 |
+
clip = ImageClip(image_path).set_duration(12).set_audio(AudioFileClip(audio_path))
|
59 |
+
clip = clip.set_fps(1)
|
60 |
+
clip.write_videofile(output_path, codec="libx264", audio_codec="aac", fps=1)
|
61 |
+
return output_path
|
62 |
+
|
63 |
+
# Unified generation
|
64 |
+
def generate_av(prompt):
|
65 |
+
audio_path = generate_audio(prompt)
|
66 |
+
image_path = generate_image(prompt)
|
67 |
+
video_path = combine_to_video(image_path, audio_path)
|
68 |
+
return video_path
|
69 |
+
|
70 |
+
# UI
|
71 |
+
interface = gr.Interface(
|
72 |
+
fn=generate_av,
|
73 |
inputs=gr.Textbox(
|
74 |
label="π€ Prompt your sonic art here",
|
75 |
placeholder="e.g. 'drunk driving with mario and yung lean'"
|
76 |
),
|
77 |
+
outputs=gr.Video(
|
78 |
+
label="π§ Generated Audiovisual Clip"
|
|
|
79 |
),
|
80 |
title='π Hot Prompts in Your Area: "My Husband Is Dead"',
|
81 |
+
description="Enter a fun sound idea for music art. Returns a synced image + audio mp4.",
|
82 |
examples=[
|
83 |
"ghosts peeing in a server room",
|
84 |
"tech startup boss villain entrance music",
|
85 |
"AI doing acid in a technofeudalist dystopia"
|
86 |
],
|
87 |
css="style.css"
|
88 |
+
)
|
89 |
|
90 |
+
interface.launch()
|