CorridorKey by blace.ai

Tips:

  • Use AE’s internal Advanced Spill Suppresor to remove bleeding colors!

  • You will get the best performance on a NVIDIA / Windows machine with Mercury GPU Acceleration (CUDA) enabled.

  • Disable Composition -> Preview -> Cache Frames When Idle for best responsiveness of the plugin.

Parameters

Hint Matte

Use either an AI-based hint matte (BiRefNet) or provide the model with a custom b/w layer, coming from another keyer.

Auto Matting Quality

The tradeoff between speed and quality in the automatting step (using BiRefNet). The “Fast” results will be sufficient in most cases.

B/W Layer

The layer used to provide the CorridorKey model with the alpha hint.

Keying Quality

This sets the internal processing resolution of the frame. “Fast” corresponds to 512px, “Accurate” to 1024px and “Extreme” to 2048px.

Output Mode

“Result” gives you the RGB output with alpha set and “Matte” the black/white matte. “Hint” is the internal matte passed to the CorridorKey model.

Backend & Performance

Hardware Acceleration (on Silicon Macs and CUDA machines only)

Run calculations on the GPU. This will give massive speedups compared to CPU mode.

Lower Precision

Compute with reduced precision if possible. This can save up to half of the memory and give you some speedups at the cost of sometimes slightly reduced quality.

CUDA Memory Sharing (on CUDA machines only)

Try to keep frames data on the GPU for rendering. This is faster, especially on larger resolutions like 4k. Might not work on some NVIDIA driver versions (e.g. 476.x) so keep your drivers updated.

Model Offloading

Enabling this will make sure only the AI model parts which are needed for computation are kept on the GPU. This might lower VRAM usage under some settings at the cost of moving AI models in and out of GPU memory. Options are:

  • No offloading: Keep all models on the GPUs VRAM.

  • CPU: Move unneeded models to the RAM and back if needed. This will occupy RAM.

  • Full Unload: Completely unload models if not needed. This saves both VRAM and RAM but might be much slower, as the models have to be loaded again for every request.

Timeout

The maximum time a frame can be processed before the computation times out.

Samples (not available for all settings)

The number of ai samples to calculate. This will improve the models accuracy.

Parallel (only available if Samples > 2)

This will render all samples at the same time (faster), if disabled computation might be slower but require less VRAM.

Computation Tiles (not available for all settings)

Split the computation into several tiles. This can help if you run out of memory.

Stop Processing

Cancels a long-running frame computation immediately.

Restart

If a frame was canceled with “Stop Processing”, trigger a rerender.