Troubleshooting
Run corridorkey doctor first. It checks the most common failure points and tells you exactly what is wrong.
Inference model not found
FAIL inference model not found in ~/.config/corridorkey/models
The model was not downloaded. Run corridorkey init and choose to download when prompted. If the download fails, see the manual download instructions printed by init.
FFmpeg not found
FAIL ffmpeg not found - install FFmpeg
FFmpeg is not installed or not on PATH.
On Windows, download from ffmpeg.org and add the bin/ folder to your PATH environment variable.
On macOS with Homebrew:
brew install ffmpeg
On Ubuntu/Debian:
sudo apt install ffmpeg
After installing, run corridorkey doctor to confirm.
corridorkey: command not found
The tool installed but the install location is not on PATH. Close and reopen your terminal. If that does not help, find where uv installed the tool:
uv tool dir
Add the path printed there to your PATH environment variable.
Clips not detected
If the wizard shows no clips or fewer clips than expected, the most likely cause is that the folder structure is not recognised.
The wizard scans for Input/, Frames/, or Source/ subdirectories (case-insensitive). If your footage is in a flat folder without these subdirectories, the wizard will offer to organise it at the start of the session. Choose y when prompted.
If you declined the organise prompt and clips are still missing, check that each clip folder contains at least one of:
Input/orinput/- image sequence or video insideFrames/orframes/- image sequenceSource/- video file- A video file named
Input.mp4(or any case/extension) directly in the clip root
Clips stuck in RAW state
my_shot RAW 60 -
RAW means the clip has input frames but no alpha hint. CorridorKey needs an alpha generator package to produce the alpha hint before inference can run. Install one and re-scan.
If you expected the clip to be READY, check that the AlphaHint/ folder exists inside the clip directory and contains the same number of frames as Frames/ or Input/.
Clips stuck in EXTRACTING state
my_shot EXTRACTING 0 0
EXTRACTING means a source video was detected but frames have not been extracted yet. This happens automatically when you choose [i] in the wizard. If the clip stays in EXTRACTING after processing:
- Check that FFmpeg is installed (
corridorkey doctor). - Check that the source video is not corrupted (try opening it in a media player).
- Check the error column in the wizard table for a specific message.
CUDA out of memory
RuntimeError: CUDA out of memory
The GPU does not have enough VRAM for the current settings. Try:
- Close other applications using the GPU.
- Reduce
refiner_scaleto 0.5 or 0.0 in the wizard settings. - Process one clip at a time rather than a large batch.
If the error persists on a 4 GB GPU, the model may not fit. CPU processing is the fallback:
corridorkey process /path/to/clips --device cpu
First frame takes very long (kernel compilation)
On the first run after installation, PyTorch compiles GPU kernels for your specific hardware. This takes approximately one minute and is normal. The wizard shows a spinner during this step. Subsequent runs use the cached compiled kernels and start immediately.
Output frames are too dark or too bright
If your input footage is linear light (EXR from a camera or renderer) and you did not enable input_is_linear, the pipeline will treat it as sRGB and the output will look wrong. Re-run with input_is_linear enabled in the wizard settings.
Matte has holes or speckles
Small holes in the matte are removed by auto_despeckle. If holes remain, increase despeckle_size in the wizard settings. If fine detail like hair is being removed, decrease despeckle_size.
For persistent edge problems, try reducing despill_strength slightly (e.g. 0.8) as aggressive despill can affect semi-transparent areas.
Config file not found warning
WARN config file not found - run `corridorkey init` to create it
This is a warning, not a failure. CorridorKey uses built-in defaults when no config file is present. Run corridorkey init or corridorkey config init to create the file.
Corrupt config file
If the config file contains invalid YAML, CorridorKey will automatically rename it to corridorkey.yaml.bak and continue with defaults. You will see a warning in the output:
WARN Corrupt config file moved to ~/.config/corridorkey/corridorkey.yaml.bak - using defaults.
Run `corridorkey config init` to regenerate.
To start completely fresh and remove all config, models, and cached data:
corridorkey reset
Then run corridorkey init to set up again.