Judging an AR Filter Like Software, Not Like a Graphic

Judging an AR Filter Like Software, Not Like a Graphic

A branded AR filter usually gets judged the way a poster gets judged: does it look good in the clip the agency sent you. That is the wrong test. A filter is not a finished image. It is a small piece of software that has to keep running on a stranger’s phone, in a stranger’s room, on a face it has never seen before.

A filter is closer to a small app than a post

A static graphic or a cut video is finished the moment it is exported. Whatever it looked like on the agency’s monitor is what everyone will see, forever. An AR filter does not get that guarantee. It has to keep working across face shapes it was never tested on, skin tones with different reflectance, a bedroom lamp instead of a ring light, an older phone doing the tracking math, and a front camera with a narrower field of view than the back camera the agency filmed the demo on.

None of that shows up in a thirty second showcase clip, recorded under the best of those conditions, on the best device, by someone who already knows exactly how far to turn their head for the tracking to lock on cleanly. That clip is evidence the idea works. It is not evidence the build does, and a portfolio will happily let you confuse the two.

Put it on your own face before you judge it

If the filter is still live, the most useful thing you can do is stop watching the demo and open the real thing on your own device, under ordinary conditions rather than the ones the demo was staged for.

  • Use your normal room lighting, not a setup arranged for the test.
  • Move at an ordinary pace, not the slow, deliberate turn demo reels rely on.
  • Watch for tracking that lags behind your face or hand by a visible beat.
  • Watch for the effect snapping off entirely when you turn too far or the light shifts, instead of degrading gracefully.
  • Watch for elements that clip through a jaw or a hand instead of sitting convincingly on top of it.

Any one of those failures, on your own phone, in your own room, tells you something the demo was built to hide.

Where the craft actually lives: the idea versus the execution

Two different questions get collapsed into one. Is the concept original, and is the thing actually built well. A filter can be a clever idea built badly, or a competent build of an idea a dozen other agencies have already shipped this year.

The weak pattern is easy to spot once you know it: an effect that is a recoloured or resized template asset with the brand’s logo dropped on top, contributing no interaction design of its own. The strong pattern puts the interaction itself at the centre: the effect responds differently depending on what the person does, a gesture, an expression, how long they hold still, rather than decorating their face the same way regardless of behaviour. Reading the person versus just watching a face go by is what separates an idea that uses AR from a piece of AR design.

Reading the reel footage for tells, not just the finished shot

Most portfolios only show a highlight reel, so it helps to know what a reel is telling you even when you cannot test the filter yourself.

A reel cut from many short clips, none longer than two or three seconds, is often hiding a filter that could not sustain a longer continuous take without breaking. Footage shot from a fixed, front on distance, with no real movement, is a tell that the effect was not built to tolerate motion or angle change. A robust build usually shows at least one clip proving it holds up off axis, because that is the shot that actually proves durability, not just the idea.

What actually counts as a stronger use of the format

Picture two versions of a filter built around the same idea: a hand motion that triggers a transformation on screen. The weaker version only works if the gesture matches the demo exactly, same distance, same speed. The stronger version still works when the gesture is done slightly wrong, because the build accounted for how a real person moves rather than the one exact motion a demo performer used.

There is a second thing worth checking that almost nobody thinks to: what happens when the person stops interacting. A considered build resolves cleanly, settling back to neutral or fading out. A rushed build tends to freeze or glitch instead, because the exit moment is usually the first thing skipped once the team is out of time. Whether an effect knows how to end is a real craft signal, not a footnote.

Where this crosses into a different question

How to build a filter in a given platform’s tool is not this site’s subject. And whether a filter “performed well” by view count or completion rate is a reported metrics question for a different kind of review, not a craft judgment. This piece is judging the object an agency built, not whether the agency should have picked this format for the brief at all.

For the broader pattern this fits inside, see what a portfolio leaves out. And it’s worth remembering, here as elsewhere, that one excellent piece tells you almost nothing on its own.

See the work worth judging

Our own selections get made the same way: by looking at the object an agency actually built, not the reel cut to flatter it. See this year’s selections, or if an agency’s interactive work deserves a closer look, nominate an agency.

FAQ

Why would a filter work perfectly in an agency’s showcase clip but fail when I try it?

Because the showcase clip is recorded on a chosen device, in controlled lighting, by someone who already knows the exact motion the tracking responds best to. None of that setup is available to you when you open the same filter cold, which is exactly why the demo and the real experience can diverge.

Is a simple filter automatically a weaker piece of craft than a complex one?

No. A simple gesture based effect that stays robust across imperfect attempts and resolves cleanly is stronger craft than a visually complex effect that only holds up in one narrow, rehearsed condition.

Can I judge a filter’s craft just from a portfolio reel, without trying it myself?

Partially. Short cuts and no off axis or extended take footage are real signals. But they are secondary to actually testing the filter directly whenever it’s still available to try.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *