Two formats dominate AI-made short-form ads. One generates scenes — B-roll, product shots, atmospheric motion — straight from a description. The other puts a person on screen reading your script. Neither is better in the abstract. The right one depends on what you're trying to make the viewer do.
The core difference
- Text-to-video builds scenes. You describe a shot and get motion: a product on a counter, a city at dusk, an abstract loop behind your text. No person required.
- Avatar-led ads build connection. A talking head — an avatar or your Personal Clone — speaks directly to the viewer, carrying the message with a face and a voice.
When text-to-video fits
Reach for text-to-video when the thing is the story, not the spokesperson. It shines for:
- Product and lifestyle B-roll — showing the item in use, in context, in motion.
- Mood and atmosphere — openers that set a tone before any words land.
- Anything wordless — visual hooks that work on mute, which is how most short-form is watched.
- Scenes you can't film — locations, scale, or setups that a shoot can't reach.
Because there's no person to anchor it, text-to-video is also the most flexible to remix. Swap the scene description and you have a fresh visual without rebuilding the whole ad.
When avatar-led fits
Reach for an avatar-led ad when trust, explanation, or a direct ask is the point:
- Explaining a benefit that needs a human voice to feel credible.
- Direct-response hooks — "Here's why you're doing this wrong" lands harder from a face.
- Founder or brand presence without filming, especially via a Personal Clone.
- Multi-language delivery — same script, same person, new language, no reshoot.
A person looking at the camera creates a parasocial pull that scenery can't. When you need the viewer to believe something, put a face on it.
You don't have to choose
The strongest ads usually use both. A common, effective structure:
- Open with a punchy text-to-video scene or B-roll hook to stop the scroll.
- Cut to an avatar delivering the core message and the offer.
- Close on product B-roll with your CTA.
In Krex AI you can compose this in one pass — talking heads, AI B-roll, and text-to-video scenes live in the same pipeline, and every render is 1 credit = 1 video, so combining formats doesn't cost extra to test. See the full toolset for how scenes and avatars stitch together.
Quick guide by goal
- Awareness / scroll-stopper: text-to-video.
- Trust / explanation / direct ask: avatar-led.
- Conversion ad with a hook and an offer: both, in sequence.
- No-camera founder content: avatar-led with a Personal Clone.
Which should you pick?
If your message lives in a visual, start with text-to-video. If it lives in a voice, start avatar-led. And if it's an actual ad meant to convert, layer them — hook with a scene, persuade with a face, close on the product.
Ready to try the mix? Spin up a few on the Free tier and see which format your audience rewards. New here? Start at the overview.