What actually makes a good jumbotron video (we made a bunch)

We've put a lotof selfies through OnTheTron — ours, friends', testers', the occasional dog. After enough of them, you start to see the difference between a clip that makes someone go “wait, send me that” and one that gets a polite “haha cool.” The gap is surprisingly consistent, and almost none of it is random. Here's what we've learned actually makes a good jumbotron video.

What separates a great clip from a flat one

1. The face reads instantly

The clips people love are the ones where you recognize the person in the first half-second. That sounds obvious, but it's the whole ballgame. When the input selfie is sharp, well-lit, and face-forward, the output is unmistakably them— and that recognition is what makes a friend react. When the input is dark, angled, or filtered, the face comes out a little generic, and a generic face kills the “that's YOU on the screen” magic. (We wrote the full input checklist in how to take the perfect selfie.)

2. The realism lives in the still, not the motion

Early on we assumed more motion = more “alive.” Wrong. The best clips have restrained motion — a little crowd shimmer, a touch of camera life — layered on a still that already looks real. Overdo the movement and faces wobble into the uncanny zone. The lesson shaped our whole two-step pipeline: get the still convincing first, then add just enough motion to sell “live broadcast.”

3. The broadcast cues do the heavy lifting

A score bug in the corner, telephoto crowd framing, stadium lighting, and a 16:9 widescreen frame all signal “this is TV” before the viewer consciously processes anything. Strip those out and even a perfectly-rendered face just looks like a person at a game. Keep them and a decent render reads as a real fan-cam catch.

What consistently flops

• Sunglasses and hats.If the model can't see your eyes, it invents them — and invented eyes are exactly where “that's not quite me” comes from. This is the single most common reason a clip disappoints.
• Tiny faces in wide shots.If your face is a small part of the original photo, there isn't enough detail to rebuild you faithfully. Fill the frame.
• Heavy beauty filters. They erase the real texture and proportions that make you you. The more “perfected” the selfie, the less recognizable the result. Counterintuitive, but true every time.
• Group photos (for now).The experience is tuned for one clear subject. Crowds of faces split the model's attention and nobody comes out sharp.

The pattern behind all of it

Almost everything above reduces to one idea: give the AI a clean, unambiguous read of your face, then let the broadcast styling do the theater. The tech is good, but it can only work with what you hand it. A great input + restrained motion + real broadcast cues = the clip that gets screenshotted into the group chat. A muddy input fighting heavy motion = the one that gets a shrug.

The fastest way to internalize all this is to just run a couple yourself: try one with a clean, well-lit selfie and one with sunglasses, and the difference is immediate. Make your first one here — it's free — or browse more examples to calibrate your eye first.