Recording

[recording] decides how a screen recording is encoded. All three keys are read when a recording starts.

Key	Type	Default	Notes
recording.quality	low | medium | high | max	high	One fixed quantizer per encoder, so a still screen costs nothing and a busy one what it needs
recording.encoder	vaapi | nvenc | software | auto	auto	What encodes H.264; auto takes the first that works, in that order
recording.device	DRM node path	unset	The node VA-API encodes on, like /dev/dri/renderD128

[recording]
quality = "high"
encoder = "auto"
device = "/dev/dri/renderD128"

NVENC is not tried while VA-API works, so an NVIDIA GPU that sleeps stays asleep. One that does not work gives way to software, which kanade doctor and the log say. An unset device finds an Intel or AMD node, the GPU that drives the display first, and never an NVIDIA one.

kanade capture record start screen system
kanade capture record stop
kanade doctor

Note

Each step of quality changes the size of a busy picture by about a quarter. A recording needs wf-recorder, plus an encoder that works; the window pipeline additionally needs gst-launch-1.0 with the PipeWire plugin and that of the chosen encoder.

See also

The Capture Surface, capture gaps, and media and capture CLI.