I’ve been going back and forth on this and would love to hear how others in the community are approaching it.
LiveKit now ships some really polished audio visualizer components (bar, radial, aura, etc.) that give an agent a visual presence with zero additional infra cost, low latency, and no uncanny valley risk. On the other end of the spectrum, the avatar plugin ecosystem — Hedra, Tavus, Simli, and the rest of the 12 providers in the docs — offers lip-synced video that ranges from stylized to hyper-realistic.
I keep landing on a few tensions I’d love other builders’ perspectives on:
-
Uncanny valley vs. trust. Has anyone found that a realistic avatar actually increases user trust and engagement in their use case, or do users respond just as well (or better) to an abstract visualization that doesn’t try to be human? I’m curious whether the answer changes by domain — e.g., customer support vs. education vs. internal tooling.
-
Cost/complexity tradeoff. Avatar providers add a separate participant to the room, another vendor dependency, and per-minute cost. For those of you shipping avatars in production, is the engagement lift clearly worth it, or is it more of a “nice demo” feature?
-
The middle ground. Is anyone experimenting with something between a pure audio visualizer and a full lip-synced face — like stylized/cartoon avatars, or visualizers that incorporate some anthropomorphic elements without going full photorealistic?
Not trying to knock avatars at all — there are clearly use cases where visual embodiment is the whole point. I’m genuinely trying to figure out where the line is for different product contexts. Would appreciate any experience reports, user testing anecdotes, or strong opinions.