revid

Simple video restoration for Python. Load a video, chain restoration filters, export. That's it.

Built for restoring VHS tapes, old camcorder footage, digitized film, and any degraded video. Uses FFmpeg for fast classic filters, with optional AI-powered models for superior quality.

Installation

pip install revid

With optional extras:

pip install revid[ai]   # AI-powered restoration (Real-ESRGAN, GFPGAN, Demucs, etc.)
pip install revid[all]  # everything

Requires FFmpeg installed and available in your system PATH.

Note on AI features: The revid[ai] extras install PyTorch and several AI models (~2GB+). AI engines process video frame-by-frame and are computationally heavy — a dedicated NVIDIA GPU with CUDA support is strongly recommended. FFmpeg-only features run on any machine with no extra dependencies. AI engines were tested on an NVIDIA RTX 5070 Ti (16GB VRAM) with PyTorch CUDA 12.8.

Usage

import revid as rv

video = rv.read("family_tape_1994.mp4")

# Chain restoration steps and export
video.deinterlace() \
     .denoise(strength=0.5) \
     .color_correct(saturation=1.2, brightness=0.05) \
     .sharpen() \
     .upscale(factor=2) \
     .to_mp4("restored.mp4")

# Inspect intermediate steps
deinterlaced = video.deinterlace()
deinterlaced.to_mp4("step1_deinterlaced.mp4")

denoised = deinterlaced.denoise()
denoised.to_mp4("step2_denoised.mp4")

# Preview a frame without rendering the full video
video.deinterlace().denoise().preview(at=5.0, output="preview.png")

# Use presets
video.preset("vhs_standard").to_mp4("restored.mp4")
video.preset("vhs_ai").to_mp4("restored_ai.mp4")

# GPU-accelerated encoding (NVIDIA NVENC)
video.preset("vhs_standard").to_mp4("restored.mp4", gpu=True)

# AI-powered upscaling
video.upscale(factor=4, engine="realesrgan").to_mp4("upscaled.mp4")

# AI face restoration
video.upscale(factor=2, engine="realesrgan") \
     .face_restore(engine="gfpgan") \
     .to_mp4("faces_restored.mp4")

# AI audio enhancement
video.audio_denoise(engine="demucs").to_mp4("clean_audio.mp4")
video.audio_separate(engine="demucs", stem="vocals").to_mp4("vocals_only.mp4")

# Video metadata
info = video.info()
print(info)

Each transformation returns a new object. The original is never modified.

Presets

FFmpeg (fast, no dependencies)

AI-powered (require revid[ai])

Engine Pattern

Every filter method defaults to FFmpeg. Use the engine parameter to switch to an AI model:

# FFmpeg (default, fast)
video.upscale(factor=2)
video.upscale(factor=2, engine="ffmpeg", algorithm="lanczos")

# AI (slower, higher quality)
video.upscale(factor=4, engine="realesrgan")

# Same pattern for all methods
video.denoise(strength=0.5)                      # FFmpeg hqdn3d
video.denoise(strength=0.5, algorithm="nlmeans")  # FFmpeg nlmeans
video.denoise(engine="nafnet")                    # AI

video.stabilize()                                 # FFmpeg vidstab
video.stabilize(engine="raft")                    # AI optical flow

video.face_restore(engine="gfpgan")               # AI only
video.colorize(engine="deoldify")                  # AI only

Features

Video Filters (FFmpeg)

Audio Filters (FFmpeg)

AI Engines (optional)

Utility

Supported Formats

License

BSD 3-Clause