Griffin-Lite access form: what to prepare while you wait

To try Griffin-Lite you request access through the form linked from Tavus's Griffin page (read 2026-10-06). The page says Griffin-Lite is a research preview for select trusted testers, that full Griffin is not yet available, and that Tavus is still building safe disclosure features. It lists no pricing, so budget decisions have to wait.

What to have ready

The page does not publish the form's questions, so this is a preparation list, not a copy of the form. Write down the one conversation you want a live face to hold, who will be on the other side, and how you will tell them it is AI. Tavus says Griffin can lead people to believe they are talking to a human, and that it is working on disclosure, so a clear disclosure plan is the part most likely to matter.

• One concrete use case with a start and an end, not a general interest in video AI.
• A plan to label the avatar as AI on screen and in speech.
• Whether the face needs your own likeness. Get consent for any real person's photo first.
• A fallback for the days you are not admitted: a rendered clip.

What you can ship this week on Sume

Sume does not offer a live, listening avatar. It renders clips. You create an avatar once with Create new avatar, from a text prompt, structured traits or a photo, through POST /v1/avatar-1.0/generate. You then call POST /v1/avatar-1.0/talking-video with that avatar's handle and a script of 4 to 60 seconds.

That covers the scripted half of the use case: the welcome, the answer to a known question, the follow-up. It does not cover a back-and-forth. Keep the two apart in your plan.

Do not wait on a form to start

A request form gives no date. Rendering a clip today gives you scripts, framing and disclosure wording that you can reuse if a live model becomes available to you later. Nothing in the clip workflow locks you in, because your scripts and photos are yours.