Tavus Phoenix-4.5 adds cartoon and anime faces; Sume's options

Tavus Phoenix-4.5 extends its face model to animated human styles: cartoon, anime and Pixar-style. Sume's avatar docs do not list styles at all. Avatar creation takes a text prompt, structured profile traits, or a reference image, so whether a stylised character works is something to test with one cheap create, not assume.

The Tavus facts come from its changelog; the Sume facts from Create new avatar and Generate avatar video.

What did Tavus ship with Phoenix-4.5?

The September 2 entry says you can start using a face in minutes from a photo or a video, after which it tunes in the background into an unwatermarked version. It adds more natural movement below the neck, allows glasses, jewelry and hair in front of the shoulders, and extends support to animated human styles (cartoon, anime, Pixar-style). Phoenix-4.5 became the default for new face training on September 9, 2026, and on September 10 Tavus added 50 first-party faces built on it, including 28 Pro faces.

Tavus's faces are built for its conversational product. The changelog does not describe rendered, script-to-video output limits, so none are claimed here.

What can Sume create an avatar from?

Three input types, each a job: prompt (describe the avatar), props (structured traits such as ethnicity, sex and age) and photo (a reference image_url). photo needs a fetchable public HTTPS image; localhost, private-network, non-HTTPS URLs and non-image responses are rejected before generation is submitted.

The handle you choose is the name you reuse. It may include a leading @, and Sume stores it without the @.

How do I test a stylised character on Sume?

Create one avatar from a short prompt or a reference drawing and read the result before building on it. A single paid create is the cheap experiment; the docs tell you to poll the job, then use the returned handle for videos.

Then check the video, not just the avatar. Create a short avatar video preview to see first-frame stills before you pay for a full render.

curl -X POST https://api.sume.com/v1/avatar-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: avatar-style-test-001" \
  -d '{
    "avatar_handle": "style_test",
    "input": {
      "type": "prompt",
      "prompt": "A friendly cartoon host with round glasses"
    }
  }'

What should I not assume?

Do not assume that because one vendor lists an animated style, another one does: Sume's docs make no promise about lip movement on non-photoreal faces. Do not treat the avatar resource as a conversational agent either. A Sume avatar speaks a script you send, in a clip of an estimated 4-60 seconds, and answers nobody.

If the first create looks wrong, change the prompt and make a new handle. Handles are how you find the avatar again, so pick a distinct one per experiment.

Which Tavus facts transfer to a Sume workflow?

The idea that a face should start quickly from a photo does transfer: a photo avatar on Sume is a single create job followed by reuse of its handle. The idea of a tuned, unwatermarked version in the background does not appear in Sume's docs, so do not expect it.

Keep your reference image clean: one public HTTPS URL that returns an image, since a page or a signed link fails the check before any spend. If the character is an illustration, say so in the prompt as well as in the reference, and compare the preview stills against your art before you approve a render.