Test for your AI agent website explainer

Your agent, as a realistic AI avatar,
walking viewers through your site.

A 33-second test made for this proposal. An AI presenter introduces himself full frame, then moves into a picture-in-picture bubble while a real screen recording shows the property search, the filters, a listing and the booking button, with zooms, callouts and a lower third.

1920 × 1080, 30 fps33 s with voice and musicLip sync to one continuous voice takeFictional agent and fictional website

What happens in the 33 seconds

Everything you asked for in one short piece: the avatar presenting, real website footage in picture-in-picture, smooth transitions, animated callouts and a clean end card.

Presenter full frame with lower third

Introagent full frame, name lower third

Search filters with callout

Searcharea, budget and bedrooms

Filtered listings with key facts callout

Listingszoom on the key facts

Booking a viewing

Listing pagephotos, facts, booking

End card

End cardagent, brand and call to action

How I’d do it with your agent’s real photos and voice

The test uses an invented presenter so no real person is copied. For your video the same pipeline runs on the material you send, with your agent’s written consent for the avatar and the voice clone.

Voice clone

From 2 to 5 minutes of clean audio of your agent I create a consent-based ElevenLabs voice clone. You approve the script, then the whole voiceover is recorded in one continuous take so it sounds natural.

Avatar from footage

From his photos and a short video I build the avatar (HeyGen, or an audio-driven lip sync model like the one in this test). I check every shot at full size: face, teeth, hands and lip sync.

Screen recordings

I record your real website in a clean browser, scripted so scrolling, filters and clicks land exactly on the words of the voiceover. Crisp 1080p, no jitter, no private data.

PiP edit

Avatar and recordings come together in a motion-graphics edit: picture-in-picture, zooms on the features, animated text and callouts, lower third, end card, music mixed well under the voice.

What you get

  • Website explainer with your agent as a realistic AI avatar and his cloned voice
  • Real screen recordings of your property search, listings and features
  • Picture-in-picture layouts, zooms, animated callouts and lower thirds in your colours
  • 16:9 master in 1080p, plus a 9:16 cut for social on request
  • Script polish included, 2 revision rounds
3 to 4 days
from receiving photos, video and voice samples

First full draft within 3 days, final delivery after your feedback. Because the edit is built in code, changes to text, timing or callouts are quick and do not touch the avatar footage.

Tools used for this test: Nano Banana Pro for the presenter still, ElevenLabs for voice and music, OmniHuman 2.0 for the lip-synced talking head, a small mock website built in HTML and screen-recorded with Playwright, Remotion for the edit. Harbour & Oak Realty, its listings and the agent are fictional; property photos are AI generated.