Rate avatar clips like Griffin-Lite: naturalness, trust, flow

The Tavus Griffin page lists average ratings of 5.4 for naturalness, 5.6 for trustworthiness and 4.9 for conversation flow from its Griffin-Lite study, with naturalness on a 7-point scale. You can borrow the first two questions to score any avatar clip you render, and skip the third, which only applies to live conversation. A 15-second clip on each Sume tier costs $14.69 in total.

The sheet

Show each clip to five colleagues who have not seen the script. Ask them to rate each item from 1 to 7, then average. Keep the clips blind: label them A, B and C, not standard, plus and max.

What the test costs on Sume

Use one script and one avatar handle. At the catalog rates, 15 seconds is $2.76 on standard, $3.675 on plus and $8.25 on max, which is $14.69 for all three. The avatar video guide lists standard as the fastest path, plus as the default and max as the highest quality with slower turnaround.

If you want to approve the composition first, the preview guide says first-frame stills are tier-independent, so one preview serves any tier you later choose for the final render.

How to read the result

Do not compare your averages with Tavus's. Their raters watched a live model in a different study, and I do not know how they sampled. Use the scale only as a template. What matters is the gap between your own tiers on your own script.

If standard and plus score within a quarter point, pay for standard. If max beats plus by a full point on trustworthiness, the extra $0.305 a second may be worth it for a hero ad. For a throwaway reply clip it almost never is.

Keep the record

Save the job id for each clip, with the tier, the script and the average score. A table of ten rows is enough to show later why you chose plus over max, and it lets a teammate repeat the test when the script changes.