This is an experimental algorithm that allows continuous, real-time reuse of an existing Talking-Head Avatar video sequence (e.g. D-ID, SadTalker). New text-input can re-use the avatar without generating a new video from the provider. The original video contains all the mouth-shape-sequences that the algo needs. Basically (with some additional magic), the algo takes the sequences that it needs and stitches them together to match the new input-text.

For any thoughts on this approach or suggested use-cases, please contact me at: feedback@helporia.com