Legal 77.3%, financial 78.4%: a review step before an avatar render

If you publish avatar videos in a regulated profession, put a written sign-off between the script and the render, and keep the job record with it. On Sume the approved text is the script or video_inputs you send to POST /v1/avatar-1.0/talking-video, so the cheapest control point is before that call, and the job result is your proof of what was rendered.

The adoption numbers are from HeyGen's The State of AI Avatars 2026, a September 2026 survey of 1,000+ small business owners: 77.3% adoption in legal services, 78.4% in financial services, 65.6% in real estate and 46.1% in health and wellness. It is a vendor survey, not a rule about what is permitted. Nothing here is legal advice, and Sume does not decide what your regulator requires.

Why does a regulated script need a gate before rendering?

Rendering is a paid, hard-to-undo step. An avatar reads whatever you send it, including a number, a claim or a guarantee that nobody reviewed. Changing a script later means a new render. The avatar video previews guide says a changed script or video_inputs needs a new preview, so a late edit costs a full second pass.

What does the review step look like?

Keep it as small as the risk allows. A table in your own tracker is enough, with one row per clip:

How do you wire the gate into the API flow?

Create a preview first, review the stills together with the script, and call generate-video on the preview id only after sign-off. The preview stage returns an avatar_video_preview_id plus job polling URLs. Stored captions on the preview apply at generate-video, and preview stills are never burned with captions.

Use a distinct Idempotency-Key per clip and write it into the tracker next to the reviewer's name. If a clip is regenerated after edits, create a new preview and a new key rather than reusing the old one.

# 1) preview first, do not render yet
curl -X POST https://api.sume.com/v1/avatar-video-previews \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: clip-0412-preview-v1" \
  -d '{"avatar_handle":"advisor_host","script":"Your plan review is booked for Tuesday at 10.","aspect_ratio":"9:16"}'

# 2) after sign-off, render from the preview id
curl -X POST https://api.sume.com/v1/avatar-video-previews/$PREVIEW_ID/generate-video \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: clip-0412-render-v1" \
  -d '{}'

What does Sume not cover?

Sume does not review content, check claims, add disclaimers, or file anything with a regulator. It does not keep your approval record either; that is yours. It also generates clips, not live conversations, so a viewer cannot ask the avatar a question. What it gives you is a reproducible request, a preview stage and job records you can point to when someone asks what was published and when.

What should the reviewer actually look at?

Read the script as spoken text, not as a document: numbers, names, dates and any promise. Then look at the preview stills for the presenter and setting, because an avatar in a setting that implies an office or credential can say more than the script does. Finally confirm the disclosure wording and where it will sit, since that is part of what the viewer sees.

Keep the reviewed text byte for byte as the request body. If the script is edited after sign-off, the review no longer applies to the render.