← The Thread: notes on shopping an outfit from a photo

How it works

Why can cropping too tightly make a search worse, not better?

Cropping usually helps. Crop past the point of context, though, and a search loses the surrounding detail it needed to tell a sleeve from a full garment.

Save to Pinterest

Cropping usually improves a search, and most of the advice on this site says so. There is a limit, though, and cropping past it removes exactly the context a search needs, turning a helpful edit into one that makes the result worse.

Why does cropping normally help?

Because it gives a garment a larger share of a photo's fixed resolution and removes unrelated background, other people and other garments that would otherwise compete for the same attention. How to crop a photo for a better match covers this general benefit in depth, and it holds true for most garments most of the time.

The benefit comes from concentrating the photo's available detail on the one thing you actually want described. That benefit has a limit, though, and the limit is where the crop starts removing information the search actually needs rather than information it does not.

What information does a search actually need beyond fabric detail?

The garment's overall silhouette and its boundaries: where it starts and ends, whether it has sleeves and how long they are, what kind of neckline or collar it has, how it falls relative to the body. These are the features that distinguish one garment category from another, and they exist at the scale of the whole garment, not at the scale of a small patch of fabric.

A crop that removes all of these while zooming in tightly on a small section of fabric gives a search excellent detail about texture and colour, and almost nothing about what kind of garment that texture belongs to.

How does this actually go wrong in practice?

A photo cropped so tightly that only a section of a sleeve is visible, with no shoulder seam, no cuff, no hem and no sense of the garment's overall shape, can be read ambiguously. That same patch of fabric could belong to a sleeve, a scarf, a trouser leg, or a section of a dress, and without the surrounding context a search has no reliable way to tell which.

This is the opposite failure from the one cropping is usually meant to solve. An uncropped photo with too much competing content can bury a garment in noise. An over-cropped photo removes so much surrounding structure that even a garment with plenty of visible texture becomes ambiguous about what it actually is.

How do you know if a crop has gone too tight?

A practical test: look at the cropped photo on its own, with no other context, and ask honestly whether you could confidently say what type of garment it shows. If it takes a moment of genuine uncertainty, whether that is a sleeve or a full top, a skirt or a dress, the crop has likely gone too far.

A more concrete version of the same test is checking whether at least one complete boundary of the garment remains visible: a full neckline, a complete hem, a full sleeve from shoulder to cuff, or a waistband to hem for trousers. A crop that preserves at least one of these while still tightening the frame around the garment overall is usually in the useful range rather than the too-tight range.

Does this apply equally to every kind of garment?

No, and it depends on how much of a garment's identity comes from its surface detail versus its overall shape. A plain, symmetric garment, such as a simple crew-neck t-shirt in a solid colour, tolerates a tighter crop reasonably well, since colour and fabric type carry most of what identifies it, and the overall shape adds comparatively little extra information.

A garment defined by an unusual silhouette, an asymmetric hem, a distinctive collar, or a specific cut, needs that silhouette to be visible to be read correctly, since the shape itself is a large part of what makes that garment identifiable rather than a generic version of its category. The tighter-crop tolerance is genuinely different between these two cases.

What is the actual balance to aim for?

A crop that removes distracting background, unrelated garments and other people, while keeping the full garment or at least one complete boundary-to-boundary view of it visible, such as neckline to hem for a top or waistband to hem for trousers. This achieves the resolution benefit that motivates cropping in the first place while preserving the silhouette information a search still needs to place the garment into the right category before it can describe the detail within that category accurately.

Can a photo be recropped if the first attempt was too tight?

Only if the original, wider photo still exists to crop again from. Once a photo has been cropped, saved as a new file, and the original discarded, whatever was outside that crop is genuinely gone and cannot be recovered through any further editing, following the same logic covered in why the same garment gets described differently in two photos about information loss being permanent rather than reversible.

Keeping the original, uncropped photo saved until you are confident the crop you have chosen actually works well is a small habit that avoids losing the ability to try again with a wider frame.

Does digital zoom count as cropping for this purpose?

Effectively, yes. Digital zoom on most phone cameras works by cropping the sensor's captured image and enlarging what remains, which is functionally the same operation as cropping a photo afterward, just performed at the moment the photo is taken rather than afterward in an editing step. The same risk applies either way: zooming in far enough during the photo itself to lose the garment's overall silhouette produces the identical problem to an overly tight crop applied to an already-taken photo.

What is the quickest way to check a crop before uploading it?

Look at the cropped image alone, away from the original photo, ideally on a larger screen than a phone if one is available, and honestly judge whether the garment's type and overall shape remain clear from that image by itself. If it takes real effort to work out what kind of garment is even shown, the crop has likely removed too much context, and stepping back to a wider frame, one that still shows a complete boundary of the garment, is worth trying instead.

Does this problem show up differently for small accessories than for clothing?

Yes, in a way worth understanding since it can seem to contradict the general advice for small items. Finding jewellery and small accessories recommends cropping close for small items specifically because there is so little fabric or material to begin with that giving it the largest possible share of the frame is almost always a net gain. The too-tight problem described here applies more to garments with a meaningful overall shape, where cropping past the point of showing that shape trades away information the garment actually needs to be identified correctly.

The difference comes down to how much of the item's identity lives in its overall silhouette versus its surface. A ring or an earring has almost no silhouette to lose, so cropping tight is close to a pure gain. A coat, a dress, or a jacket has a silhouette that is part of what identifies it, so a crop needs to preserve at least enough of that shape to keep the garment recognisable as the category it belongs to.

Is there a way to tell in advance whether a garment needs a wider crop?

A reasonable rule of thumb: if the garment could be confused with a different category entirely once only a section of its fabric is visible, such as a sleeve looking like a scarf, or a skirt panel looking like a dress panel, it needs a wider crop that includes at least one full boundary. If the garment's category is already obvious from colour and general shape alone, such as an earring or a ring, a tighter crop is safe and usually beneficial. Thinking through this before cropping, rather than cropping as tightly as possible by default, is what avoids the too-tight mistake in the first place.

What should you actually do with this in practice?

Treat cropping as a tool aimed at removing what is genuinely irrelevant, background, other people, unrelated garments, rather than a tool aimed at maximum magnification of one piece of fabric. For most garments, that means cropping to include the whole item or at least one full boundary-to-boundary view of it. For small accessories with little independent shape, cropping tighter is safe and usually the better choice. Knowing which situation you are in before you crop is what turns cropping back into the improvement it is meant to be, rather than a step that accidentally removes the exact information a search needed most.

Common questions

If cropping generally helps, why would it ever make a result worse?

Because a crop that removes the context needed to understand a garment's overall shape leaves a search with only a fragment. A tight crop showing just a section of fabric with no visible neckline, hem, sleeve or silhouette gives less usable information than a wider crop that shows the whole garment, even though the tighter crop shows the fabric detail in more resolution.

What context does a search actually need beyond the fabric itself?

The overall silhouette and the garment's boundaries, meaning where it starts and ends, whether it has sleeves, what kind of neckline or collar it has, and how it falls on the body. These features are what separate one garment category from another, and a crop tight enough to remove all of them can leave a search unable to tell a cropped photo of a sleeve from a photo of a whole fitted top.

How do you know if a crop has gone too tight?

If a person looking at the cropped photo without any other context could not confidently say what type of garment it is, whether it is a sleeve, a full top, a skirt or a dress, the crop has likely removed too much. A useful test is asking whether the crop still shows at least one full boundary of the garment, such as a hem, a neckline or a full sleeve length.

Does this apply the same way to every garment type?

No, and it depends on how much of the garment's identity comes from its overall shape versus its surface detail. A plain, symmetric garment like a simple t-shirt can tolerate a tighter crop reasonably well, since a lot of what identifies it is colour and fabric type. A garment defined by an unusual silhouette, an asymmetric hem, or a specific neckline shape needs that silhouette visible to be read correctly.

What is a good balance between cropping too little and too much?

A crop that removes distracting background, other people, and unrelated garments while keeping the full garment or at least its complete outline visible from one identifiable boundary to another, such as neckline to hem for a top, or waistband to hem for trousers. This captures the fabric and colour detail cropping is meant to improve while preserving the silhouette a search needs.

Can you always recrop a photo if the first crop was too tight?

Only if the original, wider photo is still available to crop again from. Once a photo has been cropped and saved as a new, smaller file, and the original discarded, the information outside that crop is permanently gone and cannot be recovered by any further editing. Keeping the original uncropped photo until you are confident in the crop is worth doing for exactly this reason.

Does zooming digitally count as cropping for this purpose?

Effectively, yes, since digital zoom on most phone cameras works by cropping the sensor's image and enlarging what remains, which is the same operation as cropping after the fact, just performed before the photo is saved. The same risk applies: zooming in far enough to lose the garment's overall silhouette produces the same problem as an overly tight crop applied afterward.

Is there a way to check a crop before uploading it?

Looking at the cropped image on its own, ideally on a larger screen than a phone if possible, and honestly assessing whether the garment's type and overall shape are still clear from that crop alone is the most reliable check. If it takes a moment of guessing to work out what garment is even shown, that is a sign the crop removed too much context.

Keep reading

How it works

Why do two people in a photo confuse a search?

Group photos are the most common upload that comes back wrong. The search reads every garment it can see and has no way of knowing which person you meant.

8 min read
How it works

Why does the search pick the wrong garment?

You wanted the jacket and it described the shirt underneath. That is a detection problem rather than a matching problem, and the two have completely different fixes.

7 min read
How it works

Why does lighting change what a search finds?

Colour is one of the fields Snagfit records, and light decides what colour a garment appears to be. A navy coat under warm light reads as brown, and the search goes looking for a brown coat.

7 min read

← Back to Snagfit