7 Common Virtual Try-On Problems (and What Actually Causes Them)

Most articles about virtual try-on either sell it or dismiss it. Neither is that useful if you're actually deciding whether to trust a result or add the feature to a store. What follows is the specific, named list of things that go wrong, pulled from academic comparisons, vendor postmortems, and Corlen's own engineering history, not a vague "AI isn't perfect yet" shrug.

What are the most common virtual try-on problems?

Not every problem shows up on every generation, and not every tool has all seven. But if you've seen a virtual try-on result that looked slightly off and couldn't say why, it was almost certainly one of these.

Why does the body sometimes look different than the photo?

This is the least talked-about problem and arguably the most consequential one, since it works against you exactly when you're relying on the preview to judge fit. Because most fashion photography a model learns from shows slim people in clothes tailored to fit them perfectly, some AI try-on systems quietly narrow a waist or soften how tight a garment actually sits, rather than rendering the true strain of a structured piece on a real body.

This is a documented, named problem, not just a hunch. A 2026 paper from Google Research and the University of Washington, introducing a large fit-aware dataset called FIT, states it directly: most virtual try-on models excel at showing what a garment looks like but largely ignore whether it actually fits, defaulting to a well-fitted render even when the real size relationship between the body and the garment would produce visible strain or excess fabric. A separate 2026 paper, FitVTON, built its own real-world test set specifically because existing models kept showing clean, well-fitted results on mismatched sizes instead of the true fit.

For a shopper, the practical effect is a preview that looks more flattering than reality, which is the opposite of what a fit preview is supposed to do. For a store, it means a try-on result can look good and still send back a return, because the tool quietly did the tailor's job instead of showing the actual fit.

Why does the garment look pasted on instead of worn?

This is the most common complaint shoppers actually voice, usually as "it looks fake" without being able to say why. The technical answer is almost always the same: the model treated the garment reference image as something to composite onto the photo rather than something the person is wearing. The tell is specific: sleeves that end in open air instead of at a wrist, a collar that doesn't sit against the neck, or shadows that don't fall where the body actually is.

• Worn correctly looks like: shoulder seams sitting on the shoulder line, sleeves following the person's real arm position and ending at the wrist, fabric wrapping the visible body outline with a natural shadow.
• Pasted-on looks like: garment edges that float independent of the body, no shadow interaction, a shape that doesn't change even when the person's pose clearly would change it.

This failure gets worse, not better, exactly when it matters most: on a real customer's phone photo rather than a studio shot.

Does a low-quality photo make these problems worse?

Yes, consistently. A cropped, dim, or low-resolution photo gives a model less real information about the body it needs to dress, which is exactly the condition under which pasted-garment results, scale mismatches, and body distortion all become more likely, not less. This matters because budget-phone photos are the realistic input for most shoppers, not a controlled exception to design around later.

A tight waist-up crop is a separate case: it removes the lower body from the frame entirely, forcing a model that's asked to show a full outfit to invent legs and feet it never saw. That's a different failure than the seven above, since nothing in the original photo was there to preserve.

What can a shopper or merchant actually do about it?

1. As a shopper: use a clear, front-facing, full-body photo where possible, and treat an unusually flattering result with a bit of skepticism, particularly for structured or fitted pieces like blazers and tailored jackets.
2. As a merchant: upload clean, well-lit, uncropped garment reference photos. A blurry or heavily edited product photo caps how accurate any try-on result on it can be, regardless of the try-on engine.
3. As a merchant: if you're evaluating tools, ask specifically whether the system compares against the shopper's actual body size, not just a size label, since that's the difference between an accurate preview and a recolored default fit.
4. As a merchant: build in a way to catch bad results. A tool with no feedback loop has no way to know a specific garment or photo type is consistently producing pasted-on or distorted results.

Corlen's approach to this is covered in more depth in is virtual try-on accurate, including the specific engineering steps Corlen takes: locking the face, pose, and background so only the clothing region changes, comparing garments against the shopper's real body rather than a fixed chart, and an automatic check after every generation that catches a result too close to an unedited passthrough and retries it rather than serving it. None of that makes every generation perfect. It's the difference between a tool that assumes every result is fine and one built to catch it when a result isn't.

For how these results connect to what actually reduces returns, see how to reduce online fashion returns. To see how a real generation handles your own photo, including on an ordinary phone shot rather than a studio one, try Corlen free, or read the developer docs if you're evaluating the API for your own store.