Weekly toolbox talk as a 45-second avatar clip: 52 weeks priced
A weekly 45-second safety briefing from a Sume avatar on a worksite photo, captions on: 52 clips cost $440.96 on Standard. Tiers and script rules.
A weekly toolbox talk can be a 45-second clip from one reusable avatar with your worksite photo as the scene and captions burned in, because many sites are too loud to rely on sound. Fifty-two weekly clips with captions cost $440.96 on Standard, $583.70 on Plus and $1297.40 on Max, counting the fixed $0.20 caption add-on per clip that the API reference says the estimate includes.
The mechanics are in Sume's avatar video guide, read 2026-10-04. This post is about producing the briefing, not about what a briefing must contain: your safety officer owns the content, and a generated presenter does not replace a talk with the crew.
One avatar, one scene photo
Create the presenter once with POST /v1/avatar-1.0/generate; the creation fee is $0.95. For the background pass scene as a photo reference: scene: { "type": "photo", "image_url": "https://..." }. The image must be a public HTTPS URL, and signed or private URLs are rejected before submission. Use a real photo of the area the talk is about, with no faces in it.
The cost of a year of talks
| Item | Standard | Plus | Max |
|---|---|---|---|
| One clip, no captions | $8.28 | $11.02 | $24.75 |
| One clip, captions on | $8.48 | $11.22 | $24.95 |
| Four weeks, captions on | $33.92 | $44.90 | $99.80 |
| 52 weeks, captions on | $440.96 | $583.70 | $1297.40 |
Script rules for a briefing
Keep one hazard per clip. A 45-second script can hold one hazard, one rule and one question for the crew to discuss. Avoid numbers that are not on your own site procedure, and have the safety lead read the script before it goes to the render.
- One hazard per week, named in the first sentence.
- One rule, stated as an action, not a warning.
- End on a question the supervisor asks the crew in person.
- Put the generated-presenter note in the first caption line.
Submit the weekly job
Use the week number in the Idempotency-Key so a retry after a timeout does not create a second paid job.
curl -X POST https://api.sume.com/v1/avatar-1.0/talking-video \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: toolbox-2026-w41" \
-d '{"avatar_handle": "site_safety", "aspect_ratio": "16:9", "quality": "standard", "scene": {"type": "photo", "image_url": "https://example.com/site-gate.jpg"}, "script": "This week: ladders. Check that both feet of the ladder sit on firm ground, and keep three points of contact while you climb. Supervisors, ask your crew where the nearest ladder inspection tag is.", "captions": {"enabled": true, "style": "slam", "language": "en"}}'Review before it reaches the crew
Play the clip before it is shown. Read the captions against the script, because captions.status can come back failed while the video itself is fine. Say in the clip, the caption or the page around it that the presenter is AI-generated; the API does not add that label for you.
Sources
Related posts
More in Use cases
- Meta AI disclosure: which Sume outputs count, video and audio
Meta asks for disclosure of photorealistic video and realistic audio that was digitally created or altered. A sorting guide for video, avatar and music.
- Do YouTube Shorts AI tools label automatically? What the page says
YouTube says its own Shorts generative AI tools disclose automatically. Regional limits per tool, and what an API pipeline still has to disclose itself.
- Xiaohongshu note video: 1:1 or 9:16, 500 MB, 720p
A third-party guide to Xiaohongshu note video specs lists 9:16 or 1:1, up to 15 minutes, up to 500 MB and 720p minimum.
- Yahoo Japan Brand Panel video: 16:9, 200 MB, 5 to 60 seconds
Reported Yahoo Japan Brand Panel video specs and how to hit 16:9 and 5-60 seconds with Sume: pick a 16:9 clip, stitch past the per-job cap, and probe the file.
Written by Sume