The Thread: notes on shopping an outfit from a photo

How it works

How does a filter change what a search finds?

A filter is a colour edit someone chose on purpose, added on top of whatever the room's own light already did. Snagfit reads the finished picture, filter and all.

Save to Pinterest

A filter is not the room's light. It is a decision someone made after the photo was already taken: pick a preset, apply it, post it. That decision changes colour and contrast on top of whatever the room's own light had already done, and Snagfit reads the picture that comes out the other end of both.

Why does lighting change what a search finds? covers the ambient side of this, the light that was actually in the room. This post is about the edit made on purpose afterward, and why the two are worth telling apart even though both end up changing a colour word in a result.

What does a filter actually do to a photo?

It shifts three things across the whole image at once: hue, saturation and contrast.

Hue is the actual colour, and a filter usually pushes every hue in the photo a small distance in the same direction: warmer toward orange, cooler toward blue, or toward whatever tint the preset is built around. A navy jacket does not stay navy under a strong warm filter. It moves toward brown, the same direction every other colour in the frame moved.

Saturation is how intense a colour looks, separate from which colour it is. A high-saturation filter makes every colour more vivid, which can push a muted olive green toward a colour that reads as a completely different shade. A desaturated filter does the reverse, flattening colours toward grey.

Contrast is the gap between the lightest and darkest parts of the image. A high-contrast filter deepens shadows and brightens highlights, which can lose detail in a garment's darker folds the same way underexposure does. A low-contrast filter flattens everything toward the middle, which can make a black garment read as dark grey.

A filter applies all three at once, in a fixed combination someone designed on purpose, uniformly across the whole photo.

Why does a filter confuse a colour search?

Because Snagfit's colour field is built from the pixels it is actually given, and a filter changes those pixels before Snagfit ever sees them.

The search does not know a filter was applied. It reads a colour, records that colour as one of the fields describing the garment, and searches shopping listings for something matching that colour word. If the filter shifted navy to brown, the description says brown, and the search looks for brown.

Nothing failed here in the way a bug fails. The description is accurate about the photo it was given. The photo is an inaccurate account of the garment, because a filter deliberately made it one.

How is this different from what ambient lighting does?

The two overlap in effect and differ in cause, and the cause matters for what you can do about it.

Ambient light is uneven and specific to one room or one moment: a warm bulb here, a shadow there, daylight fading as the sun moves. It was never chosen for the photo, and it varies from one frame to the next even in the same location.

A filter is applied as one fixed transformation across the entire image, chosen deliberately by whoever posted it. That uniformity is actually useful: a filter's effect is at least consistent across every pixel in the frame, where uneven room light is not. Knowing a strong warm filter was used tells you the whole photo shifted the same direction, which is more than uneven lighting usually tells you.

The two stack rather than cancel. A photo taken under warm indoor light and then run through a warm-toned filter pushes a garment's colour further in the same direction than either change alone would. A cool filter applied to the same warm-lit photo can partly correct it, though by an amount that is not predictable just by looking.

What does a vignette do to a garment search?

A vignette darkens the corners and edges of a photo to pull attention toward the centre, and it is a common part of a filter preset rather than a separate effect someone adds on its own.

The problem is not the darkening as a visual style. It is that any part of a garment sitting inside that darkened edge loses the detail Snagfit would otherwise read from it. A sleeve cuff, a hem, or a shoe positioned near the corner of a vignetted frame can end up too dark to contribute anything to the description, the same way a shadow in poor ambient lighting would hide it.

A person looking at the photo mentally corrects for the vignette, because they expect the edges of a filtered photo to be darker and they know what a sleeve looks like anyway. Snagfit reads what is actually recorded in those pixels, which in a heavily vignetted corner can be close to nothing.

What happens with a black and white or heavily desaturated filter?

This is the clearest version of the problem, because there is no ambiguity about what got lost.

Colour is one of the fields Snagfit records for every garment, and a black and white filter removes colour from the photo entirely. There is nothing left to read. The search still runs on the fields that remain: garment type, cut, visible texture, silhouette. It runs without the field that usually narrows a result the most, which is why a black and white photo often returns a wider, less specific set of results than the same outfit shot in colour would.

A heavily desaturated filter is the softer version of the same thing. Some colour survives, but the difference between similar shades shrinks, so a result can land on the wrong one of two colours that would have been easy to tell apart in the original.

Does this apply to a colour grade on a video the same way?

Yes, and it is worth naming as the same mechanism under a different name.

A colour grade is the video editing equivalent of a filter: someone, usually the person or team who made the video, applies a colour treatment to the finished footage, often to set a mood for a scene. It shifts hue, saturation and contrast the same way a photo filter does, and it is just as deliberate.

Snagfit reads whichever frame it is given, filtered photo or graded video still, the same way: as the actual colour values sitting in that file. Whether the edit was called a filter or a grade makes no difference to what happens next.

What can you actually do about it?

Two things, and both work by giving Snagfit something closer to the garment's real colour.

Find the unfiltered original, if one exists. The same outfit is often posted more than once, or appears in a different frame of the same video shot before the filter was applied. An unfiltered version gives Snagfit colour values that were not deliberately shifted, which usually produces a more accurate colour field without you doing anything else.

Correct the colour by hand when there is only one photo. If a strong warm filter is obviously in play and the result comes back as brown or orange, treat that colour word with some suspicion. The garment type, cut and material in the description are still doing accurate work, since a filter changes colour and contrast rather than shape. Searching the same description again with the colour word swapped to what you remember, or what the un-filtered version of a similar colour would look like, often closes the gap the filter opened.

Where neither option gets you the exact colour, what to do when you cannot find the exact item covers building a search from the parts of the description a filter did not touch, which is most of them.

Common questions

Does a filter change the colour Snagfit records for a garment?

Yes, because Snagfit reads the finished photo, and a filter has already changed the colour values in that photo before it ever reaches the search. A warm filter pushes every colour toward orange and yellow, a cool one pushes toward blue, and Snagfit records whatever colour survives the edit. A navy jacket under a warm filter can be described as brown, and the search then looks for a brown jacket.

What is the actual difference between a filter and bad lighting?

Lighting is a fact about the room the photo was taken in, and nobody chose it on purpose for the photo. A filter is a deliberate edit applied afterward, on a phone or in an app, that shifts colour and contrast the same way regardless of what light the room had. Both change the colours Snagfit sees, but a filter is applied uniformly across the whole image after the fact, which makes it more predictable to reverse than uneven room lighting.

Can a filter make a search miss a garment completely?

It can, mainly through a vignette rather than the colour shift itself. A vignette darkens the corners and edges of a photo to draw the eye to the centre, and if part of a garment sits in that darkened edge, the detail there can be too dark for Snagfit to read. A sleeve or a hem sitting in a heavily vignetted corner can disappear from the description even though it is visible to a human eye that expects the darkening.

Why do warm filters cause more trouble than cool ones?

Because warm filters push toward the same colour family that warm indoor lighting already pushes toward, so the two effects stack instead of cancelling out. A photo taken under a warm bulb and then run through a warm-toned filter can shift a garment's colour further than either change would alone. A cool filter on the same warm-lit photo can partly cancel the shift, though the amount of cancellation is not something you can predict by eye.

Does a black and white filter break a colour search?

It removes colour information entirely, which is the most complete version of this problem. Snagfit cannot record a colour that is not in the photo, so a black and white or heavily desaturated filter forces the description to rely on shape, cut and visible texture instead. The search still runs, but it runs without one of the fields that usually narrows the result the most.

How is a filter different from a colour grade on a video?

They do the same thing through different tools. A colour grade is applied by whoever edited the video, often to set a mood for a scene, and a filter is applied by the person posting a photo, usually through a preset in an app. Both shift hue, saturation and contrast in the finished file, and Snagfit reads the result of either one the same way, as the actual colour values in the image handed to it.

Should you try to find the unfiltered version of a photo?

Yes, when one exists. An unfiltered original gives Snagfit colour values closer to what the garment actually looked like, which usually produces a more accurate colour field in the result. If the outfit appeared elsewhere without the filter, in a different post or a different frame of the same video, that version is worth searching instead of the edited one.

What should you do if you only have the filtered photo?

Note the filter's likely effect and correct the colour by hand afterward. If the photo has an obvious warm cast, treat a result described as brown or orange with suspicion and try the search again with the colour word adjusted toward what the garment probably was before the filter. The cut, material and garment type in the description are usually still accurate, since a filter changes colour and contrast rather than shape.

Sources

Keep reading

How it works

Why does sheer and mesh fabric confuse a search?

Sheer and mesh fabric let the camera see through the garment, into whatever is underneath. Snagfit then has to guess where one layer ends and the next one starts.

6 min read
How it works

What does an AI see in an outfit photo?

Five fields per garment, eleven possible categories, and a search query four to eight words long. Here is the exact shape of what comes back, with a real example.

7 min read

← Back to Snagfit