20-second AI welcome clip: $3.68 to $11.00 on Sume, Griffin-Lite waits

A prepared 20-second welcome message from a Sume avatar costs $3.68 at standard, $4.90 at plus and $11.00 at max (as of 2026-10-09). The Tavus page for Griffin-Lite says it is not available for customers at this time and is open to select trusted testers (read 2026-10-09), so a recorded clip is the option you can ship now.

The options side by side

The welcome is the same words every time, which suits a recorded clip. A live full-duplex model reacts to the viewer; a recorded one does not.

Fabric or avatar video

Fabric needs your own audio, up to 10 MB hosted on Sume; the avatar video generates the voice from your script. Fabric costs $3.75 here and gives you control over the voice recording; the standard avatar video is $0.07 cheaper. The choice is who supplies the voice, not the price.

A checklist for the recorded version

Write the 20-second script as three short sentences: who the avatar is, what the viewer can do next, and where to go. At a conversational pace, 20 seconds is about 2 to 3 sentences, so resist adding a fourth. If the estimator puts you over 20 seconds, trim; a 24-second clip at standard is 24 x 0.184 = $4.416, which is $0.736 more.

Choose the aspect ratio by where the clip will play. 9:16 is the default; 16:9 suits a website hero. Keep one avatar handle for every welcome so the face is the same in every language version, and render each language as its own job.

Prepare for later

The Tavus page says the company is working on safe disclosure features before broader release. Whatever you ship, plan the line that tells viewers the person is an AI avatar, in the first seconds or in the caption. Store the script, the avatar handle and the job id so you can replace the recorded clip with a live one later without reworking the content.

A welcome clip is also a good test of the multi-language case. The same avatar handle can speak a script in each language you support, one job per language, and each costs 20 x the tier rate for 20 seconds. Five languages at standard is 5 x 3.68 = $18.40.

• Write the script at 2 to 3 spoken sentences for 20 seconds.
• Render at standard first; upgrade only if the face is the issue.
• Keep the live-use wish list in the same document.