Live stream avatar: queue pre-rendered reaction clips instead

If you want an avatar on a stream that appears to react to events, pre-render a bank of short reaction clips and trigger them from your overlay software. Ten 6-second clips cost $14.70 at plus, $11.04 at standard, or $33.00 at max, with no product image. Sume avatars are async jobs: you submit a request, get a job id back, and read the finished video later. There is no live video session.

That means the avatar cannot hear your chat and answer in the moment. It can play the right prepared clip at the right moment, and for alerts such as a new follower, a milestone, or a raid, that is usually all a stream needs.

What you can build

Think of the clips as a lookup table. The event is the key and the clip is the value. A follower alert plays "welcome", a donation plays "thank you", a goal reached plays "we did it". Each clip is a normal Sume avatar job, so the minimum length is 4 seconds, and 6 seconds is a good length for an overlay that should not block the screen.

Rendering for an overlay

Pick the aspect ratio your overlay slot uses: 1:1, 3:4, 9:16, 4:3 or 16:9. Resolution is 720p at this time. Plan the clip as a framed window in your overlay. Add a scene prompt that matches your set so the clip looks like it belongs.

Because the clips are files, you can name them by event, host them yourself after downloading from the job result, and trigger them with whatever overlay tool you already use. Nothing about the playback needs Sume at all once the render is done.

Where this falls short

A bank is finite. If viewers see the same six-second clip fifty times, it stops being charming. Rotate several variants per event, and treat a bank refresh as a regular task: ten new variants at standard cost $11.04. If your show depends on a character that answers viewers in free-form speech, that is a live conversational avatar, and Sume does not provide one. A text-to-speech voice and a lip-sync clip made on demand is possible, but it takes a job's worth of time, so it works for a delayed reply and not an instant one.

Render one clip

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: stream-welcome-v1" \
  -d '{
    "avatar_handle": "stream_host",
    "script": "Welcome in, glad you are here!",
    "quality": "standard",
    "aspect_ratio": "1:1"
  }'

Planning the bank

Start with the events that happen often and matter to your community, and write three variants for each. Keep each line under about 15 words so a 6-second clip is not rushed. Use the same avatar handle throughout so the character stays recognizable, and the same scene prompt so the look matches your stream. Name the files by event and variant, such as follow-2.mp4, so your overlay rules are readable.

Keep your own record of the cost of each refresh. A bank of 30 clips at 6 seconds at standard is $33.12, which is a small, predictable cost to plan around. Each clip is a job with an idempotency key, so a script that re-runs does not duplicate a paid render when you reuse the same keys.

Disclosure

Say on your channel that the character is an AI avatar and that its lines are prepared in advance, so no one thinks it is answering them live. A clear line in your about section is enough, and it prevents the kind of confusion that a fast reaction clip can cause.

A streamer who refreshes ten clips every month at standard spends $11.04 a month, or $132.48 over twelve months. Doubling the bank to twenty variants for variety doubles that figure and no more, because cost is linear in seconds. The number to watch is not the render, it is how often you refresh. A clip that is stale after a week costs more than one that stays fresh for a quarter.