Virtual try-on
How to inspect an AI clothing try-on for visible errors
Compare a generated try-on with its inputs, inspect garment and person details, and decide whether the image is useful for your question.
Inspect a virtual try-on by comparing it with the original person photo and garment image before judging the style. Check whether it changed details that matter to your question. A realistic-looking result can still be unsuitable for a specific decision.
You are checking visible consistency, not certifying that an image is accurate. An inspection cannot prove physical fit, fabric feel, or how the outfit will move.
Decide what the result needs to preserve
Name your question first. If you are exploring a collar shape, a changed collar is a central problem. If you are comparing a print, altered pattern placement may make the comparison unusable.
Keep both inputs open beside the output. Memory is a poor substitute for a direct comparison of small details. Use similar viewing size where possible without assuming that images provide a measurable physical scale.
Do not begin by choosing the most appealing result from several. First establish whether each candidate still represents the thing you intended to compare.
Inspect garment identity
Look at the outline, neckline, sleeve length, hem, closures, pockets, and any defining pattern. Check what was visible in the input and what the result appears to invent.
A hypothetical jacket review might find:
| Detail | Input | Output | Decision |
|---|---|---|---|
| Front closure | Two visible buttons | Different closure | Unreliable for this detail |
| Sleeve | Full-length | Shortened | Do not judge sleeve style from it |
| Color impression | Dark green | Green under changed lighting | Inspect actual garment separately |
| Back | Not shown | Appears in output | Treat as inferred, not confirmed |
Apply the same detail-by-detail check to the garment you selected, focusing on the features that led you to choose it.
Compare the person and surroundings
Check whether the output changes recognizable features, body outline, pose, or unrelated parts of the scene. A changed body outline should not become evidence that the garment would alter your real shape.
Look around garment edges and overlapping areas. If hands, hair, bags, or layers become inconsistent, note whether the issue affects the question you are trying to answer.
You do not have to diagnose which technical component caused an error. “The cuff no longer matches the source” is enough to explain why you are setting the image aside.
If the discrepancy concerns the person rather than the garment, respond to a try-on that changes your appearance without treating the output as a body assessment.
Make a use-or-reject decision
A minor background discrepancy may be irrelevant to an exploratory color idea, while a changed print may invalidate it. Choose according to the task, not a universal pass score.
If the central feature is wrong or hidden, reject the result for that decision. You may try another permitted input that follows the service's requirements, but a nicer second image does not prove that it is correct.
For a size decision, use the actual product information and an appropriate physical assessment. The virtual try-on fit article explains the limit in more detail.
Use an “unable to assess” result when the source itself is unclear. If a reference hides the back of a jacket, you cannot verify a generated back detail by comparing it with that same reference. Mark the detail unknown and consult another permitted real view if it matters.
Keep visual inspection separate from a score. Finding no obvious error in a quick check does not establish that the garment, person, or physical dimensions are accurate. You may simply have no visible basis for rejecting the image on the limited question you checked. Record the question and any remaining uncertainty rather than calling the output verified. This leaves room to use a preview for inspiration without turning a casual inspection into a certification.
Only after the relevant details pass your inspection should you compare two usable try-on results for a narrow style preference.
Keep a short note if reporting a problem
Describe the input detail, the visible change, and why it mattered. Use the product's actual support or feedback channel; do not assume that every app has an in-image reporting button.
Share only the information needed and permitted for the report. Keep generated images labeled as generated if they leave the tool. The useful outcome of inspection is a clear boundary on what this particular image can help you consider.