Update app.py
Browse files
app.py
CHANGED
|
@@ -1,11 +1,12 @@
|
|
| 1 |
import torch
|
| 2 |
from diffusers import StableVideoDiffusionPipeline
|
| 3 |
from PIL import Image
|
|
|
|
| 4 |
import os
|
| 5 |
|
| 6 |
-
HF_TOKEN =
|
| 7 |
|
| 8 |
-
#
|
| 9 |
pipe = StableVideoDiffusionPipeline.from_pretrained(
|
| 10 |
"stabilityai/stable-video-diffusion-img2vid",
|
| 11 |
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
|
@@ -16,20 +17,45 @@ pipe = StableVideoDiffusionPipeline.from_pretrained(
|
|
| 16 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 17 |
pipe = pipe.to(device)
|
| 18 |
|
| 19 |
-
# === 2. Load local input image ===
|
| 20 |
-
input_path = "input.jpg" # ← Put your image in the same folder as script
|
| 21 |
-
img = Image.open(input_path).convert("RGB")
|
| 22 |
-
img = img.resize((576, 320))
|
| 23 |
|
| 24 |
-
|
| 25 |
-
|
|
|
|
| 26 |
|
| 27 |
-
#
|
| 28 |
-
|
| 29 |
-
for i, f in enumerate(frames):
|
| 30 |
-
f.save(f"frames/frame_{i:03d}.png")
|
| 31 |
|
| 32 |
-
#
|
| 33 |
-
|
| 34 |
|
| 35 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import torch
|
| 2 |
from diffusers import StableVideoDiffusionPipeline
|
| 3 |
from PIL import Image
|
| 4 |
+
import gradio as gr
|
| 5 |
import os
|
| 6 |
|
| 7 |
+
HF_TOKEN = None # Uses your Space token automatically
|
| 8 |
|
| 9 |
+
# Load pipeline once at startup
|
| 10 |
pipe = StableVideoDiffusionPipeline.from_pretrained(
|
| 11 |
"stabilityai/stable-video-diffusion-img2vid",
|
| 12 |
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
|
|
|
| 17 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 18 |
pipe = pipe.to(device)
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
+
def generate_video(inp_img, num_frames):
|
| 22 |
+
if inp_img is None:
|
| 23 |
+
return "No image uploaded!", None
|
| 24 |
|
| 25 |
+
# Resize image to SVD expected size
|
| 26 |
+
img = inp_img.convert("RGB").resize((576, 320))
|
|
|
|
|
|
|
| 27 |
|
| 28 |
+
# Generate frames
|
| 29 |
+
frames = pipe(img, num_frames=num_frames).frames[0]
|
| 30 |
|
| 31 |
+
# Save frames to video
|
| 32 |
+
os.makedirs("frames", exist_ok=True)
|
| 33 |
+
for i, f in enumerate(frames):
|
| 34 |
+
f.save(f"frames/frame_{i:03d}.png")
|
| 35 |
+
|
| 36 |
+
# Output video filename
|
| 37 |
+
out_path = "output.mp4"
|
| 38 |
+
|
| 39 |
+
# Build MP4 video
|
| 40 |
+
os.system(f"ffmpeg -y -framerate 10 -i frames/frame_%03d.png {out_path}")
|
| 41 |
+
|
| 42 |
+
return out_path
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
# Gradio UI
|
| 46 |
+
with gr.Blocks() as demo:
|
| 47 |
+
gr.Markdown("# 🐱 AI Image → Video Generator (SVD)")
|
| 48 |
+
gr.Markdown("Upload an image and generate a short AI video using **Stable Video Diffusion img2vid**.")
|
| 49 |
+
|
| 50 |
+
with gr.Row():
|
| 51 |
+
inp_img = gr.Image(type="pil", label="Upload an input image")
|
| 52 |
+
num_frames = gr.Slider(4, 24, value=8, step=1, label="Number of Frames")
|
| 53 |
+
|
| 54 |
+
btn = gr.Button("Generate Video")
|
| 55 |
+
|
| 56 |
+
out_vid = gr.Video(label="Generated Video")
|
| 57 |
+
|
| 58 |
+
btn.click(generate_video, inputs=[inp_img, num_frames], outputs=out_vid)
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
demo.launch()
|