UGC ad script for AI avatar videos: hook, demo, CTA

A UGC ad script is the spoken lines and beats of an ad made to look like a customer filmed it: a hook in the first seconds, the product in use, and a call to action. For an AI avatar on Sume, write each beat as a short scene and keep the whole ad within an estimated 4 to 60 seconds. In current code, Sume times speech at about 2.8 words per second, and a beat without words lasts the seconds you give it.

The limits come from Generate avatar video and the Sume API reference, read on 2026-09-27; the word rate and the English-only rule are read from Sume's current code. Each beat becomes one entry in the request's video_inputs; the multi-scene avatar video API covers those fields and rules, and AI UGC ad generator API compares this route with a catalog UGC Format.

What is a UGC ad script made of?

Three beats, in the order the viewer meets them. The example in Sume's docs follows this shape: a one-line hook, four seconds without words labeled demo, then the call to action, set in "Casual bedroom framing, native UGC lighting". Treat it as a starting structure, not a rule:

How long should each line be?

Budget in seconds. In current code, a spoken line is planned at the duration you set or, without one, at its word count divided by 2.8, rounded up to a whole second; a wordless beat is planned at the seconds you give it. Sume then packs the beats into clips of 4 to 12 seconds and adds those up, so the planned total can come out above the plain sum of the beats. The whole ad must land at 4 to 60 seconds.

• A 3-second hook is roughly eight words at 2.8 words per second, and a 5-second call to action roughly fourteen.
• Silent seconds count toward the same 60-second limit, so a longer demo leaves less time for speech.
• Script for an AI avatar explains why the limit isn't an exact word count.

How do I write the hook?

Open on the most surprising or useful thing the viewer is about to see, in one line they can take in at once; the docs example opens on a question. Keep it short: at 2.8 words per second, three seconds is about eight words. Each version of an ad is its own avatar video job, so trying several openings means one job per hook; hook variations for UGC ads covers that.

What can't a UGC avatar script do?

• Speak another language. In current code the clip prompt says "Spoken language: English only." For other languages, the speech goes through TTS 1.0 with a language and VEED Fabric 1.0 lip sync instead.
• Cast two people. Current execution supports one resolved avatar per final video, so a reply from a second creator is a separate job, as in AI avatar conversation videos.
• Change rooms between beats. Current execution expects scene backgrounds to resolve to one shared scene.
• Run past 60 seconds in one video. Split a longer ad into several jobs.
• Carry captions on the talking-video route. In current code, POST /v1/avatar-1.0/talking-video refuses a captions field with 400 invalid_request, even though the docs list it; send captions on an avatar video preview instead, where they apply at its generate-video step, or caption the finished clip with the video captions API.