Tavus Griffin-Lite is not available: what to render today

No, you cannot use Tavus Griffin-Lite as a customer today. Tavus's own Griffin page says Griffin-Lite is a research preview available only to "select trusted testers" and "will not be available for use for customers at this time." There is no API for it either. If you have a talking-avatar deliverable due, the practical path is a model you can call now: Sume Avatar 1.0 renders a script-driven talking video of 4 to 60 seconds from a ready avatar.

This post separates what Griffin is from what you can ship, so you can plan around the gap instead of waiting on it.

What Tavus says Griffin-Lite is

Tavus describes Griffin as a "Human Interaction Model": a full-duplex video-to-video system that models the conversation and generates audio and video together. The preview works from a single reference photograph. That is a live-conversation product, not a render-a-clip product.

The company states that its existing API still uses earlier models (Phoenix, Raven and Sparrow), and that full release is expected only after safety procedures are addressed. Pricing is not given on the page.

What a rendered clip is for

Sume Avatar 1.0 is a different shape of product. You create a reusable avatar once, then submit a script and get back a finished MP4 from an asynchronous job. It does not hold a conversation, listen, or react to a viewer. It produces the clip you would have asked a presenter to record.

That fits the cases where a live model is overkill or unwanted: a landing-page welcome, a product explainer, a UGC-style ad, a lesson intro. The same script always yields a clip you can review before it goes live.

Ship a clip now

The request needs a ready avatar handle and exactly one of script or video_inputs. Sume estimates the video length from the script and accepts 4 to 60 seconds. quality is standard, plus (the default) or max, aspect_ratio defaults to 9:16, and resolution is currently 720p, per the Sume avatar video docs.

Send an Idempotency-Key so a retry does not bill twice, then poll the job.

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: welcome-clip-001" \
  -d '{
    "avatar_handle": "product_host",
    "aspect_ratio": "9:16",
    "quality": "plus",
    "script": "Hi, I am the host of this page. Here is what you get in the first minute."
  }'

Decide between live and rendered

Pick a live model when the viewer must talk back and the avatar must answer in the moment. Pick rendered clips when the words are known in advance. Many teams need both: a rendered clip for the page, a live agent for support.

When Griffin or any similar model reaches general availability with an API, compare it on the same terms you would use today: duration limits, resolution, how a job is billed, and whether the output URL stays valid. Until then, a clip you can call is worth more than a demo you cannot.

• Check the vendor page for a customer-availability line before you design around a preview.
• Keep the script and avatar handle in version control so you can re-render on another model later.
• Review the first frame with an avatar video preview before paying for a full render.