How to take the perfect selfie for an AI jumbotron video
The single biggest factor in whether your OnTheTron video looks amazing or slightly off isn't our AI — it's the photo you feed it. We've watched a lot of selfies go through the pipeline, and the gap between a great result and a “hmm, that kind of looks like me” result is almost always the input. Here's what consistently works, what consistently doesn't, and why.
Quick context for why this matters: our two-step pipeline first re-renders your selfie as a stadium still, then animates it. If the model can't clearly read your face in the original, every step afterward inherits that uncertainty. Garbage in, slightly-uncanny out.
The five things that matter most
1. Light your face evenly
Soft, even light is the cheat code. Face a window, step under shade outdoors, or use a room with diffuse overhead light. Avoid harsh side light (half your face in shadow) and avoid strong backlight (a bright window behind you turns your face into a silhouette). If you can see your own features clearly in the photo, so can the model.
2. Face the camera, head-on
A roughly straight-on angle gives the AI the most information about your face. Extreme tilts, looking-away shots, and dramatic up-the-nose or down-from-above angles all increase the odds of identity drift. You don't need to be rigid — just point your face generally at the lens.
3. Fill a good chunk of the frame
Your face should be a clear, central part of the photo, not a tiny head in the corner of a wide landscape shot. If your face is only a few hundred pixels across, there simply isn't enough detail to reconstruct you faithfully. A normal arm's-length selfie is perfect.
4. Keep your face unobstructed
Sunglasses are the most common result-killer we see — the model can't place eyes it can't see, so it invents them, and invented eyes are where “that's not quite me” comes from. Hats pulled low, hair across the face, and masks have the same effect. Clear eyes, clear face.
5. Skip the heavy beauty filters
Light, natural photos beat heavily filtered ones. Aggressive beauty/smooth filters erase the exact micro-detail (skin texture, real proportions) that makes the output recognizably you. Ironically, the more “perfected” the selfie, the less like yourself the AI version tends to look. A plain, un-retouched shot is your friend here.
A few smaller things that help
- One person per photo. Today the experience is built around a single subject. A solo selfie gives the cleanest result.
- A calm background.You don't need a studio backdrop, but a busy, cluttered background gives the model more to untangle. A relatively plain wall or open space is easiest.
- Decent resolution. A normal phone photo is plenty; a tiny, heavily-compressed screenshot of a screenshot is not.
- Natural expression.A relaxed smile or a neutral look both work great. You're going on the big screen — a little personality reads well.
The mistakes we see most
If your result came out a little off, it was almost certainly one of these: sunglasses or a hat hiding the eyes, a dark or backlit photo, a face that's small in a wide shot, or a heavily filtered selfie. The good news is that all of them are free to fix — just retake the photo with the rules above and run it again.
Want to put it to the test? Grab a clean, well-lit selfie and make your jumbotron moment. If you're curious what “good” looks like first, browse a few real examples to calibrate.
Related reading
- Why fan-cam videos go viral on TikTok (and how to make yours pop)The fan-cam look hijacks a feeling we all recognize — being seen, celebrated, caught on the big screen. Here's why that format travels so well, and what we've noticed makes one actually take off.
- The most iconic jumbotron moments of all timeProposals, dance-offs, kiss-cam chaos, the lone superfan — a tour of the big-screen moments that became part of sports culture, and what each one teaches about why the format works.
- What actually makes a good jumbotron video (we made a bunch)After generating a lot of these, the difference between a clip that lands and one that flops is surprisingly consistent. Here's what works, what falls flat, and the patterns behind both.