Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,119 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import cv2
|
| 3 |
+
import torch
|
| 4 |
+
import numpy as np
|
| 5 |
+
import gradio as gr
|
| 6 |
+
from PIL import Image
|
| 7 |
+
from diffusers import StableVideoDiffusionPipeline
|
| 8 |
+
|
| 9 |
+
# =========================
|
| 10 |
+
# CONFIG
|
| 11 |
+
# =========================
|
| 12 |
+
MODEL_ID = "stabilityai/stable-video-diffusion-img2vid"
|
| 13 |
+
OUTPUT_DIR = "output"
|
| 14 |
+
|
| 15 |
+
NUM_FRAMES = 25
|
| 16 |
+
FPS = 7
|
| 17 |
+
SEED = 42
|
| 18 |
+
|
| 19 |
+
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
| 20 |
+
DTYPE = torch.float16 if DEVICE == "cuda" else torch.float32
|
| 21 |
+
|
| 22 |
+
os.makedirs(OUTPUT_DIR, exist_ok=True)
|
| 23 |
+
|
| 24 |
+
# =========================
|
| 25 |
+
# LOAD MODEL (ONCE)
|
| 26 |
+
# =========================
|
| 27 |
+
pipe = StableVideoDiffusionPipeline.from_pretrained(
|
| 28 |
+
MODEL_ID,
|
| 29 |
+
torch_dtype=DTYPE,
|
| 30 |
+
variant="fp16"
|
| 31 |
+
)
|
| 32 |
+
|
| 33 |
+
pipe.to(DEVICE)
|
| 34 |
+
pipe.enable_model_cpu_offload()
|
| 35 |
+
pipe.enable_attention_slicing()
|
| 36 |
+
|
| 37 |
+
generator = torch.Generator(DEVICE).manual_seed(SEED)
|
| 38 |
+
|
| 39 |
+
# =========================
|
| 40 |
+
# SAVE VIDEO FUNCTION
|
| 41 |
+
# =========================
|
| 42 |
+
def save_video(frames, path, fps):
|
| 43 |
+
h, w, _ = frames[0].shape
|
| 44 |
+
video = cv2.VideoWriter(
|
| 45 |
+
path,
|
| 46 |
+
cv2.VideoWriter_fourcc(*"mp4v"),
|
| 47 |
+
fps,
|
| 48 |
+
(w, h)
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
for frame in frames:
|
| 52 |
+
frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
|
| 53 |
+
video.write(frame)
|
| 54 |
+
|
| 55 |
+
video.release()
|
| 56 |
+
|
| 57 |
+
# =========================
|
| 58 |
+
# MAIN FUNCTION
|
| 59 |
+
# =========================
|
| 60 |
+
def images_to_video(images):
|
| 61 |
+
if images is None or len(images) == 0:
|
| 62 |
+
return None
|
| 63 |
+
|
| 64 |
+
output_videos = []
|
| 65 |
+
|
| 66 |
+
for idx, img in enumerate(images):
|
| 67 |
+
image = Image.fromarray(img).convert("RGB")
|
| 68 |
+
image = image.resize((1024, 576))
|
| 69 |
+
|
| 70 |
+
with torch.autocast(DEVICE):
|
| 71 |
+
result = pipe(
|
| 72 |
+
image,
|
| 73 |
+
num_frames=NUM_FRAMES,
|
| 74 |
+
generator=generator
|
| 75 |
+
)
|
| 76 |
+
|
| 77 |
+
frames = [
|
| 78 |
+
(frame * 255).astype(np.uint8)
|
| 79 |
+
for frame in result.frames[0]
|
| 80 |
+
]
|
| 81 |
+
|
| 82 |
+
output_path = os.path.join(OUTPUT_DIR, f"video_{idx}.mp4")
|
| 83 |
+
save_video(frames, output_path, FPS)
|
| 84 |
+
|
| 85 |
+
output_videos.append(output_path)
|
| 86 |
+
|
| 87 |
+
return output_videos
|
| 88 |
+
|
| 89 |
+
# =========================
|
| 90 |
+
# GRADIO UI
|
| 91 |
+
# =========================
|
| 92 |
+
with gr.Blocks(title="Image to Video AI") as demo:
|
| 93 |
+
gr.Markdown("""
|
| 94 |
+
## 🎥 Image to Video AI (Multi Image)
|
| 95 |
+
Upload **multiple images**, otomatis jadi **video MP4**
|
| 96 |
+
""")
|
| 97 |
+
|
| 98 |
+
image_input = gr.File(
|
| 99 |
+
file_types=["image"],
|
| 100 |
+
file_count="multiple",
|
| 101 |
+
label="Upload Images"
|
| 102 |
+
)
|
| 103 |
+
|
| 104 |
+
video_output = gr.Video(label="Generated Video")
|
| 105 |
+
|
| 106 |
+
generate_btn = gr.Button("🚀 Generate Video")
|
| 107 |
+
|
| 108 |
+
def process(files):
|
| 109 |
+
images = [np.array(Image.open(f.name)) for f in files]
|
| 110 |
+
videos = images_to_video(images)
|
| 111 |
+
return videos[0] if videos else None
|
| 112 |
+
|
| 113 |
+
generate_btn.click(
|
| 114 |
+
fn=process,
|
| 115 |
+
inputs=image_input,
|
| 116 |
+
outputs=video_output
|
| 117 |
+
)
|
| 118 |
+
|
| 119 |
+
demo.launch()
|