Tavus Phoenix-4.5: photo rules vs a Sume photo avatar

Tavus Phoenix-4.5, announced September 2, 2026, lets you start using a face from a photo or video within minutes, and allows glasses, jewelry and hair in front of the shoulders. Sume creates an avatar from a photo too, via a fetchable public HTTPS image_url, and renders scripted talking videos rather than live conversations.

Tavus claims are from its changelog; Sume's from Avatars and Generate avatar video, read 2026-09-30.

What does the Phoenix-4.5 entry say?

The face is usable in minutes, then tunes in the background into an unwatermarked version. Movement below the neck is more natural; glasses, jewelry and hair in front of the shoulders are allowed; animated human styles (cartoon, anime, Pixar-style) are supported. It became the default for training a new face on September 9, 2026.

What does a Sume photo avatar take?

The avatar create call takes an input of type photo with an image_url. The docs require a fetchable public HTTPS image URL; localhost, private-network, non-HTTPS and non-image responses are rejected before generation is submitted. The docs do not list the Phoenix-style rules above (glasses, jewelry, cartoon styles), so test your image rather than assume.

What do I get back?

Avatar videos turn a ready avatar into a script-driven talking video, with aspect_ratio of 1:1, 3:4, 9:16, 4:3 or 16:9. To approve a first frame before spending on a full render, create a preview first; see avatar video first-frame previews. For the conversation-versus-clip choice, read real-time avatar vs video avatar API.

How should I test a tricky photo?

Create the avatar from the photo, then run a short script through a preview and look at the still. If glasses or a hairstyle render badly, swap the photo and try again. Do the same for stylized art, since the Sume docs make no promise about it.