Outage status update video with an AI avatar: script and cost

Yes, a short avatar clip can sit on a status page next to the written update, but only as a second layer. A rendered clip is an asynchronous job, so it will not be ready in the first minutes of an incident. Publish the text update first, then attach the clip when it finishes. The clip is for customers who would rather watch than read a timeline.

Sume's talking-video route takes a script and a ready avatar handle and returns an MP4. The docs say the estimated duration must fall between 4 and 60 seconds, and that video-style jobs usually do not finish inside the synchronous wait window, so plan for async polling or a webhook.

A four-line script that fits 25 seconds

Status clips fail when they ramble. Keep one sentence per job below, and read it aloud once before you submit.

• What is affected, in the customer's words (not the internal service name).
• Who is affected and who is not.
• What your team is doing right now, in one verb.
• When the next update will be posted, as a clock time and time zone.

Cost by tier

Rates come from Sume's fixed avatar-video price table: standard 0.184, plus 0.245 and max 0.55 dollars per second without a product image. Quality defaults to plus when you omit it. Standard is the fastest path, which is the right pick for time-sensitive notices, and the difference between tiers on a 25-second talking head is mostly polish, not content.

What an avatar clip should not do in an incident

It cannot answer questions, and it cannot be re-recorded in seconds. If the situation changes after you render, the clip is wrong and a new job is the only fix. Treat the clip as a snapshot, stamp the time in the script and in the page copy, and take it down or replace it when the incident closes.

Do not use an avatar to apologise for something serious in a way that reads as a person. Say in the page copy that the video is AI-generated. The EU transparency checklist in this post covers the disclosure side.

Set it up before you need it

Create the avatar once ahead of time (creation is a one-time 0.95 dollars), keep the handle in your runbook, and pre-write two script skeletons. During an incident you then change a few words and submit one request with an Idempotency-Key so a double click does not create two jobs. If a job fails, the reservation is released; the failed-job post shows what the ledger does.

Where to host it and how to word the page

Put the clip under the written update on the status page, not instead of it. Give the page a line such as 'Video summary, AI-generated, posted 14:05 UTC', so that readers can tell when the snapshot was made. Search engines and screen readers cannot use a video, and some customers are on locked-down networks that block media, so the text update stays the primary record.

After the incident, keep the clip only if it is still true. A post-incident review is a better place for a longer explanation, and that can be a second, separate job if you want a presenter for it. If the review runs past a minute, split the script into two jobs; the avatar route will not accept a single script over 60 seconds.

Wording that ages well

Write times as absolute clock times with a zone and avoid 'soon' or 'shortly'. Name the symptom customers see, such as failed logins, instead of the internal cause. Leave out root-cause guesses: the clip is rendered once and cannot be corrected, while the written update can be edited as facts arrive. If the root cause is later confirmed, put it in the text post-incident review and not in a re-rendered clip.