RIFE / app.py
1inkusFace's picture
Update app.py
7ec6c48 verified
raw
history blame
8.59 kB
import spaces
import os
import shutil
import subprocess
import glob
import importlib.util
import gradio as gr
# --- Constants ---
BASE_DIR = os.getcwd()
RIFE_DIR = os.path.join(BASE_DIR, "Practical-RIFE")
MODEL_URL = "https://huggingface.co/hzwer/RIFE/resolve/main/RIFEv4.26_0921.zip"
# --- Setup Functions ---
def run_command(command):
subprocess.run(command, shell=True, check=True)
def setup_environment():
"""Sets up RIFE, downloads weights, and patches libraries."""
print("--- Starting Environment Setup ---")
# 1. Clone Repo
if not os.path.exists(RIFE_DIR):
print("Cloning Practical-RIFE...")
run_command(f"git clone https://github.com/hzwer/Practical-RIFE {RIFE_DIR}")
# 2. Download Weights
if not os.path.exists(os.path.join(RIFE_DIR, "HDv3")):
print("Downloading RIFE v4.26 model weights...")
zip_path = os.path.join(RIFE_DIR, "RIFEv4.26_0921.zip")
run_command(f"wget -O {zip_path} {MODEL_URL}")
run_command(f"unzip -o {zip_path} -d {RIFE_DIR}")
# Fix directory structure (User Fix)
train_log_dir = os.path.join(RIFE_DIR, "train_log")
os.makedirs(train_log_dir, exist_ok=True)
extract_folder = os.path.join(RIFE_DIR, "RIFEv4.26_0921")
# Move files safely
if os.path.exists(os.path.join(extract_folder, "RIFE_HDv3.py")):
shutil.move(os.path.join(extract_folder, "RIFE_HDv3.py"), train_log_dir)
if os.path.exists(os.path.join(extract_folder, "IFNet_HDv3.py")):
shutil.move(os.path.join(extract_folder, "IFNet_HDv3.py"), train_log_dir)
with open(os.path.join(train_log_dir, "__init__.py"), 'w') as f: pass
hdv3_dir = os.path.join(RIFE_DIR, "HDv3")
os.makedirs(hdv3_dir, exist_ok=True)
if os.path.exists(os.path.join(extract_folder, "flownet.pkl")):
shutil.move(os.path.join(extract_folder, "flownet.pkl"), hdv3_dir)
shutil.rmtree(extract_folder)
if os.path.exists(zip_path): os.remove(zip_path)
# 3. Patch skvideo (Numpy Fix)
try:
spec = importlib.util.find_spec('skvideo.io.abstract')
if spec and spec.origin:
with open(spec.origin, 'r') as f: content = f.read()
if 'np.float' in content:
run_command(f"sed -i 's/np.float/float/g' {spec.origin}")
run_command(f"sed -i 's/np.int/int/g' {spec.origin}")
except Exception as e:
print(f"Warning: skvideo patch failed: {e}")
# 4. Patch RIFE inference for libx264
inference_script = os.path.join(RIFE_DIR, "inference_video.py")
if os.path.exists(inference_script):
with open(inference_script, 'r') as f: content = f.read()
if "libx264" not in content:
new_content = content.replace("-c:v', 'mpeg4', '-qscale:v', '1'", "-c:v', 'libx264', '-preset', 'medium', '-crf', '23'")
with open(inference_script, 'w') as f: f.write(new_content)
print("--- Setup Complete ---")
setup_environment()
@spaces.GPU(required=True)
def interpolate_video(input_video_path, multi_factor):
if input_video_path is None: return None
factor = str(multi_factor).replace("x", "").strip()
output_path = os.path.join(BASE_DIR, "output_rife.mp4")
final_output_path = os.path.join(BASE_DIR, "final_interpolated.mp4")
# Clean previous runs
if os.path.exists(output_path): os.remove(output_path)
if os.path.exists(final_output_path): os.remove(final_output_path)
# RIFE script creates a specific output name if audio transfer fails
# It often appends _noaudio, so we must watch for that.
expected_no_audio = output_path.replace(".mp4", "_noaudio.mp4")
if os.path.exists(expected_no_audio): os.remove(expected_no_audio)
os.chdir(RIFE_DIR)
try:
print(f"Running RIFE with {factor}x on {input_video_path}")
cmd = ['python3', 'inference_video.py', '--video', input_video_path, '--output', output_path, '--multi', factor, '--model', 'HDv3']
subprocess.run(cmd, check=True)
# --- Logic to handle the file finding ---
# If moviepy fails, RIFE creates 'output_rife.mp4' OR 'output_rife_noaudio.mp4'
# depending on where it crashed. We check for both.
source_to_encode = None
if os.path.exists(output_path):
source_to_encode = output_path
elif os.path.exists(expected_no_audio):
source_to_encode = expected_no_audio
print("Audio transfer failed inside RIFE, using no-audio version.")
else:
print("Error: Output video file not found after inference.")
return None
# Re-encode for web compatibility
print(f"Re-encoding {source_to_encode} to {final_output_path}...")
subprocess.run(['ffmpeg', '-i', source_to_encode, '-c:v', 'libx264', '-pix_fmt', 'yuv420p', '-movflags', '+faststart', '-y', final_output_path], check=True)
return final_output_path
except Exception as e:
print(f"Error during interpolation: {e}")
return None
finally:
os.chdir(BASE_DIR)
# --- Stitching Logic ---
def stitch_videos(video_files, resolution_choice):
if not video_files: return None
# Parse resolution
try:
target_w, target_h = resolution_choice.split("x")
target_w, target_h = target_w.strip(), target_h.strip()
except:
target_w, target_h = "1920", "1080"
print(f"Stitching {len(video_files)} videos into {target_w}x{target_h}...")
stitch_list_path = os.path.join(BASE_DIR, "stitch_list.txt")
output_stitched = os.path.join(BASE_DIR, "final_stitched.mp4")
temp_dir = os.path.join(BASE_DIR, "temp_stitch")
if os.path.exists(temp_dir): shutil.rmtree(temp_dir)
os.makedirs(temp_dir)
# 1. Normalize
normalized_files = []
for i, vid_path in enumerate(video_files):
temp_filepath = os.path.join(temp_dir, f"norm_{i}.mp4")
# Scale and Pad Filter
scale_filter = f"scale={target_w}:{target_h}:force_original_aspect_ratio=decrease,pad={target_w}:{target_h}:(ow-iw)/2:(oh-ih)/2"
cmd = [
'ffmpeg', '-i', vid_path,
'-r', '60',
'-vf', scale_filter,
'-c:v', 'libx264', '-crf', '23',
'-c:a', 'aac',
'-ar', '44100',
'-y', temp_filepath
]
subprocess.run(cmd, check=True)
normalized_files.append(temp_filepath)
# 2. Create List
with open(stitch_list_path, 'w') as f:
for path in normalized_files:
f.write(f"file '{path}'\n")
# 3. Concatenate
print("Concatenating...")
subprocess.run(['ffmpeg', '-f', 'concat', '-safe', '0', '-i', stitch_list_path, '-c', 'copy', '-y', output_stitched], check=True)
shutil.rmtree(temp_dir)
return output_stitched
# --- Gradio Interface ---
with gr.Blocks() as demo:
gr.Markdown("# 🎞️ RIFE: Interpolate & Stitch")
with gr.Tabs():
# TAB 1: Interpolate
with gr.TabItem("1. Smooth Video (Interpolate)"):
gr.Markdown("Upload a single video to increase its framerate.")
with gr.Row():
with gr.Column():
video_input = gr.Video(label="Input Video")
multi_select = gr.Dropdown(["2", "4", "8"], value="2", label="Multiplier")
interp_btn = gr.Button("Interpolate", variant="primary")
with gr.Column():
video_output = gr.Video(label="Smoothed Output")
interp_btn.click(interpolate_video, inputs=[video_input, multi_select], outputs=video_output)
# TAB 2: Stitch
with gr.TabItem("2. Stitch Videos"):
gr.Markdown("Upload multiple videos. They will be normalized to **60fps**.")
with gr.Row():
with gr.Column():
stitch_inputs = gr.File(label="Upload Clips", file_count="multiple")
res_select = gr.Dropdown(choices=["1920x1080", "1280x1280", "1024x1024"], value="1920x1080", label="Resolution")
stitch_btn = gr.Button("Stitch Videos", variant="primary")
with gr.Column():
stitch_output = gr.Video(label="Stitched Result")
stitch_btn.click(stitch_videos, inputs=[stitch_inputs, res_select], outputs=stitch_output)
demo.launch()