Mac & Linux

kukkiri is a Python package with a command-line tool and an optional desktop app.

Requirements

NVIDIA GPU, compute capability 7.5+, driver 580+
Python 3.10+
ffmpeg on the PATH

pip

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130   # match your CUDA
git clone https://github.com/treehog2/kukkiri.git && cd kukkiri
pip install .

For the desktop app: pip install ".[gui]", then run kukkiri-gui.

Docker

Needs the NVIDIA Container Toolkit.

docker build -t kukkiri .
docker run --rm --gpus all -v "$PWD:/data" -v "$PWD/weights:/weights" kukkiri \
    restore /data/input.mp4 -o /data/restored.mp4 --weights /weights

Keep the weights folder mounted writable so the TensorRT engines built on the first run are reused.

Mac

Mac support isn't available yet. kukkiri currently needs an NVIDIA GPU, which Macs don't have.

Usage

kukkiri restore input.mp4 -o restored.mp4

`kukkiri restore movies/ extra.mkv -o restored/`	Batch: files and folders (searched recursively) into an output folder
`--start 37:45 --duration 30`	Restore only a section
`--max-1080p`	Output 4K sources at 1080p: faster, smaller
`--skip-check`	Restore even if the input looks uncensored
`kukkiri download-weights [folder]`	Fetch the weights ahead of time

Weights are cached in ~/.cache/kukkiri/weights, or the folder in KUKKIRI_WEIGHTS / --weights.