A local video avatar combines components with different roles: voice, portrait animation, lip synchronisation and editing. A moving photograph is not automatically a complete presenter. For a business, the starting point is the required material: an animated portrait, a spoken message or a demonstration showing the presenter alongside a product.

Separate the voice from the speaker’s image

A narrator can record the script, or TTS (text-to-speech) can synthesise its spoken version. Select the voice model separately from portrait animation and lip synchronisation. A lip-sync component receives audio and adapts the mouth region. Animation can control head movement or expressions. The editor assembles those results with demonstrations, copy and the company’s visual identity.

This separation also helps with corrections. Fix a mispronounced name at the voice stage; fix an unsuitable cut in editing. Retain the approved script, audio and visual material separately so changing a sentence follows a clear revision path.

Components of a video presenter

  1. 1

    Voice

    Reviewed text → authorised recording or synthesis.

  2. 2

    Face and movement

    Portrait plus audio or motion reference → animation.

  3. 3

    Editing

    Presenter, demonstrations and copy → final material.

What MuseTalk and LivePortrait address

MuseTalk is an audio-driven lip-synchronisation project. It does not independently generate the script, voice and entire scene. Evaluate its processed facial region and mechanism against the required framing. Results and speeds published by the authors remain tied to their configurations.

LivePortrait focuses on motion-driven portrait animation. Its inputs differ from a text-to-avatar system. Before choosing it, describe the movement to transfer and the elements that should remain stable: identity, gaze, background and framing. Then identify the project and components suited to that operation.

Person working on a laptop in an office with artwork on the wall
Mikhail Nilov
Select a component for the required output
NeedComponentImportant input
NarrationRecording or TTSText and authorised voice
Synchronised lipsLip-sync such as MuseTalkFinal audio and compatible facial material
Animated portraitAnimation such as LivePortraitPortrait and motion reference
Advertising presenterComplete workflow and editorScript, image, voice and product assets

The photograph and the person’s permission belong in the project

For a real person’s portrait, establish image and voice use, channels and approved materials. Permission to use a photograph in an article does not automatically describe using that person as a synthetic advertising presenter. Retain that distinction in project records and asset selection.

The reference photograph should show the face clearly with suitable lighting. Check accessories, hair, facial edges and position changes throughout the sequence. The stock photograph in this article illustrates working context; it does not identify the person as a WebHat avatar or a model output.

Review the complete licensing chain

MuseTalk permits commercial use of its own code and model while requiring checks of auxiliary components. Its demonstration materials have separate conditions. A video from the project’s demonstration does not automatically become available for a company advertisement, even when the application supports commercial use.

Dependencies in the LivePortrait ecosystem can include InsightFace models. InsightFace code is MIT-licensed, while models distributed by the project have separate conditions, including non-commercial research terms. The implementation record must identify every facial model, version and permission before it enters a commercial workflow.

Component and asset records

  1. 1

    Person

    Image, voice and authorised purpose.

  2. 2

    Software

    Code, main models and auxiliary models.

  3. 3

    Delivery

    Approved material and agreed channels.

Build a script for business communication

An illustrative example explains appointment stages: what information a customer submits, what confirmation they receive and where useful details appear. The presenter can explain while the real interface demonstrates the action. Approve names, deadlines and conditions before preparing the voice.

A message claiming a customer’s personal experience differs from explaining a service. Define the company avatar’s presenting role concretely and use authorised demonstrations. For changing information, such as prices and availability, retain editable text separately from the spoken image.

Review the full sequence with and without sound

Check pronunciation, mouth closure during pauses, identity, head movement and transitions between sentences. Listen separately to product names, numbers and abbreviations. For bilingual material, check each Romanian and English track, including after the last correction.

Select hardware for the exact component and required resolution. A speed demonstrated on a server GPU does not describe a laptop’s performance. Retain audio, inputs, the project and identified models at handover.

Review a spoken message in order

  1. 1

    Text

    Names, numbers, conditions and meaning approved.

  2. 2

    Audio

    Pronunciation and pauses checked.

  3. 3

    Video

    Synchronisation, identity and editing reviewed.

Consultancy for a practical workflow

WebHat uses customised local AI and can help define the components and review steps. AI integration in a custom application can organise scripts, assets and approvals while retaining responsibility for each stage.

Practical questions

Does a lip-sync model also produce the voice?

In this workflow, the final voice is an input prepared separately.

Can any photograph be animated?

Check rights, the person’s permission and the component’s technical requirements.

Does an MIT licence cover every model?

Check code licensing separately from weights, dependencies and demonstration materials.

What should be compared in Romanian and English?

Meaning, pronunciation, pauses and synchronisation in each version, serving the same editorial purpose.