Tavus Griffin-Lite is invite-only: what can you build today?

If you cannot get into Tavus Griffin-Lite, build today with rendered clips: Sume's Avatar 1.0 talking video turns a script and a reusable avatar into a 4 to 60 second clip, and the lip-sync and motion-control routes cover still-plus-audio and still-plus-motion jobs. Griffin-Lite is a different product class, a live two-way conversation model, and Sume does not ship one.

On 2026-10-01 a model tracker listed Tavus Griffin-Lite as an invite-only research preview of a full-duplex video-to-video conversation model (Digital Applied tracker, read 2026-10-06). The tracker line gives no price or release date, and that is all this post relies on. Everything below separates what that preview is from what you can call from your own server this week.

What does an invite-only preview mean for a shipping plan?

An invite-only preview has no contract you can plan around. The tracker gives no list price, rate limit or date, so any roadmap that depends on it needs a fallback that works now. Tavus does sell a developer product today: its pricing page lists a free Basic plan with 25 minutes of conversational video, Starter at $59 a month with 100 minutes, and Growth at $397 a month with 1,250 minutes (Tavus pricing, read 2026-10-06). Those plans are for conversations, which is a live session in which the avatar listens and answers.

If your product needs the avatar to react to a person in real time, Sume is the wrong tool and you should wait for or apply to a live product. If your product needs a presenter who delivers a prepared message, a rendered clip is the right shape, and nothing about it depends on an invite.

What ships on Sume for a talking head today?

Three surfaces cover the talking-head cases. All of them are asynchronous jobs: you submit, get a job id, and read the result when it is done.

How do you start without waiting on anyone?

Create an avatar once with POST /v1/avatar-1.0/generate from a prompt, a profile or a reference photo, then reuse its handle in every clip. The Avatar video docs describe quality as standard, plus (the default) or max, and the response gives you a job to poll. If you want to approve the first frame before paying for the full render, the preview flow lets you regenerate stills and then call generate on the preview id.

curl -X POST https://api.sume.com/v1/avatar-1.0/talking-video \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: griffin-fallback-001" \
  -d '{
    "avatar_handle": "product_host",
    "aspect_ratio": "9:16",
    "quality": "standard",
    "script": "Here is what changed in this week release, in twenty seconds."
  }'

When should you revisit the question?

Re-check the tracker and Tavus's own pages when Griffin-Lite gets a public price and a date. Until then, decide on the interaction model, not the brand. A live conversation needs a streaming session and a vendor that offers one. A prepared message, a localized variant, a weekly episode or a product explainer needs a render, and a render can be re-run, captioned, trimmed and cached, which a live session cannot. For the live side of that comparison, see live avatar conversation versus an avatar job and what Griffin changes.

One practical point: write the script so that it would also work as a spoken answer. If Griffin-Lite or a similar product reaches you later, a clean script and a ready avatar still image carry over, and you will only have swapped the delivery channel.