What Sume Avatar 1.0 Does and Does Not Do vs Live Avatars

Sume Avatar 1.0 renders a scripted talking video from a reusable avatar. It does not hold a live conversation, listen to a viewer or react in real time. If you need a presenter who answers questions, this is not that product; if you need a clip you can review before anyone sees it, it is.

For context, Tavus's Griffin page, read on 2026-10-05, describes Griffin-Lite as a research preview of a full-duplex model that outputs 720p video in 320 ms chunks with 0.43 s audio-to-video latency on H100s. The page says it is not available to customers and is open only to select trusted testers, and it mentions no pricing or API.

What does it do today?

The table lists what Sume's own docs and rate card state, each read on 2026-10-05. Per-second prices are for the 720p output, without a product image.

What does it not do?

It does not run a real-time or live conversation, and Sume makes no such claim. It does not listen to a viewer, take turns, or react to what someone says mid-clip. It does not render beyond 60 seconds in one job, so a longer script has to be split into several jobs, per Generate avatar video.

It also does not give you a measured duration in advance: the 4 to 60 second check uses Sume's estimate from the script, at 2.8 words per second rounded up per scene.

When should you pick which?

Choose a scripted render when the message is fixed: onboarding, ads, reminders, notices. Review a first-frame preview, render once and send it to everyone.

Choose a live model only if a viewer must be able to interrupt. At the time of reading, Griffin-Lite is a research preview, so there is nothing to buy yet. Check Tavus's page for changes, and keep your scripted clips as the dependable path. Disclosure matters either way: Tavus reports 48 percent of 54 study participants believed they were talking to a real person, so label an avatar as one.