Cutout removes your background without a green screen and drops you into an animated, branded scene. You can record the result.
It uses MediaPipe's selfie segmenter, loaded as a float16 TFLite model through the Tasks Vision ImageSegmenter. Segmentation labels every pixel as person or background in each video frame, and that mask decides which pixels are kept. It runs on the CPU delegate, because the GPU one returned empty masks on some machines.
Edges can wobble around hair and fast hands, as with any small on-device model.