AI Video Restoration & Enhancement Setup Guide
Best for: No local GPU, fast setup, on-demand usage
Sign up at runpod.io and add credits to your account.
Select an RTX 4090 (the 4090 template will work with 3090 as well) pod (or any NVIDIA GPU with CUDA 12.1+ (except Blackwell 50xx or newer)).
mandachain/video-restoration-4090:v1 mandachain/video-restoration-5090:v1Once the pod starts, click Connect, then click the Interface link. The interface will load automatically — upload your video and start restoring.
Best for: Own NVIDIA GPU, privacy, unlimited usage
Download and install Docker Desktop. On Windows, make sure WSL 2 backend is enabled (Docker will prompt you during setup).
Docker needs GPU access. Install the latest NVIDIA Container Toolkit and make sure your NVIDIA driver is up to date (CUDA 12.1+ required). On Windows with Docker Desktop + WSL 2, GPU pass-through works out of the box with a recent driver — no extra toolkit install needed inside WSL.
Open a terminal and pull the pre-built image: (the 4090 image will work on 30/40-series GPUs, the 5090 image is for 50-series GPUs)
docker pull mandachain/video-restoration-4090:v1
or docker pull mandachain/video-restoration-5090:v1
Start the container with GPU access and port mapping:
docker run --gpus all -p 7860:7860 mandachain/video-restoration-4090:v1
ordocker run --gpus all -p 7860:7860 mandachain/video-restoration-5090:v1
To keep restored videos accessible on your host machine, mount an output folder:
docker run --gpus all -p 7860:7860 -v C:\Videos\restored:/app/outputs mandachain/video-restoration-4090:v1
ordocker run --gpus all -p 7860:7860 -v C:\Videos\restored:/app/outputs mandachain/video-restoration-5090:v1
Once the container is running and Gradio has loaded, open your browser to http://localhost:7860. Upload your video and start restoring.
video-restoration-4090:v1 image has been tested on RTX 4090 via RunPod but should work with any NVIDIA RTX 30-series or 40-series GPU (e.g. RTX 3080, 3090, 4070, 4080). You need at least ~8 GB VRAM for 720p and ~12 GB+ for 1080p output. Performance will vary by card.
docker run --gpus all nvidia/cuda:12.1.0-base-ubuntu22.04 nvidia-smi.