Free Intro Class
    Botspot Logo

    Menu

    LoginSign Up

    agenthub · output.avatar

    Avatar Video

    Animate an avatar image so it speaks — a lip-synced talking-head clip from text or from audio you supply.

    category Outputin text · audio_url · message · source_urlout —execution asyncmodel cost model tokens

    Why the node earns its place

    The talking-head path, kept separate from Generate Video because it needs a subject: a source image, animated. Give it text and it speaks; give it an audio URL and it lip-syncs to that instead.

    How it works at run time

    1. 1
      Take the text to speak, or an audio URL to sync to, plus the avatar image.
    2. 2
      Call the video capability to produce the clip.
    3. 3
      Emit the same video marker Video Output uses, so one surface renderer covers both.

    What you wire

    Inputs

    • text
      string
    • audio_url
      string
    • message
      object
    • source_url
      string

    Outputs

    Nothing handed onward.

    Configuration

    KeyTypeDefault & optionsWhat it does
    model *stringchoices from model-library:video_modelsTalking-avatar (video) model to synthesize with
    source_urlstring—Public URL of the avatar image to animate (the face that speaks). Must be reachable by the provider — not a data-URI or a localhost URL. A clear, front-facing headshot works best.
    voice_providerstring—Voice provider for spoken text when no audio is wired (D-ID text→speech), e.g. 'microsoft' or 'elevenlabs'. Blank = default.
    voice_idstring—Voice id for the selected voice provider (e.g. Microsoft 'en-US-JennyNeural'). Blank = the provider's default voice.

    What usually goes wrong

    Watch for this

    The source image must be a public URL the provider can actually reach — not a data URI and not localhost. A clear, front-facing headshot works best; anything else tends to produce an uncanny result rather than an error you can act on.

    Behaviour & provenance

    buildersVideo workflows
    routesno
    side effectsnone — computes only
    replay safetynot applicable
    talks toa model
    holdsworkspace runtime token