The two required inputs
The engine needs a video source with a detectable face and an audio source. The audio may come from BOLEME AI's voice engine or another lawful source. The face can be live, virtual-camera material or recorded content.
- Video input: camera, capture device, virtual camera, character or recorded presenter.
- Audio input: AI voice, human voice or an authorized recording.
- Output route: a tested Windows live-production chain suitable for the target platform.
What affects real-time quality
Face angle, occlusion, lighting, resolution, frame rate, audio clarity, language rhythm, hardware performance and downstream streaming software can all change the result.
Use the final face material and final audio route when evaluating quality. A short offline sample does not prove the latency or stability of a production live room.
Live person, character or recorded material
A live person can remain on camera while approved audio drives the mouth movement. Character or digital-human material can use the same audio-driven principle. Recorded presenter material can also be reused when the rights and intended use are clear.
Each source has different trust, production and monitoring requirements; there is no single setup that fits every live room.
Rights and human control remain essential
The person, face, voice, music and video must belong to the operator or be covered by valid permission. A human should be able to stop, correct or replace the output if the face tracking, audio or live route fails.
Frequently asked questions
Does the lip-sync engine write the live script?
No. Its job is to drive visible mouth movement from input audio. Script generation, product knowledge, voice production and live operations are separate responsibilities.
Can it work with a real camera?
A real-camera face is one supported input pattern, subject to face visibility, lighting, hardware and end-to-end testing.
Can it reuse recorded presenter material?
Recorded material can be part of the workflow when the face and video are authorized and the final output is tested for the intended use.