How Fast Should Virtual Try-On Be? Why Generation Speed Matters

Virtual try-on has spent most of its short history being judged on one thing: does the result look real. Speed got treated as a secondary concern, something to fix once the image quality was good enough. That is changing. As output quality across several vendors levels out, the seconds between a shopper's tap and a finished image are becoming a real point of competition, not an afterthought.

How long does virtual try-on generation actually take?

There is no single industry-standard benchmark that times every vendor under identical conditions, so the numbers below come from what each vendor states publicly or from documented product behavior, not a controlled test run in this article.

A claim of 30 to 60 seconds isn't automatically outdated marketing, either. Some garment categories, higher output resolutions, and multi-garment combinations take longer to render than a single top on a plain background, regardless of vendor.

Why does generation speed matter to a shopper?

A shopper who taps "try this on" is at the exact moment they decide whether to keep engaging or drift to another tab. Every second between that tap and a visible result is a second where attention can go somewhere else, especially on a phone, where nobody is standing next to them keeping the moment alive the way a sales associate would in a fitting room.

• In-store kiosk: a customer standing at a tablet feels every extra second physically, in a way a shopper scrolling at home on their own schedule does not.
• Mobile browsing: most virtual try-on activity happens on a phone, where a slow render competes with every other app one swipe away.
• Multi-item sessions: a shopper trying five outfits pays the wait five separate times, so even a two-second difference compounds across a single visit.

Genlook's own reported figures make the stakes concrete: according to the company, shoppers across its more than 600 live stores who try a garment on add it to cart three times more often than shoppers who do not. That lift only shows up if the shopper actually waits to see the result, which is exactly the part a slow generation puts at risk.

Does a faster result mean a worse one?

Not automatically, but the relationship between speed and quality is not a straight line either. A faster inference path can be just as accurate as a slower one; the model architectures behind Nano Banana Pro and dedicated try-on models already trade different strengths against each other well before speed enters the picture.

Watch for a shortcut that trims steps specifically to hit a faster number. That tends to show up as a garment that looks pasted onto the photo rather than worn on the body, or a background that drifts between the original photo and the result. Ask a vendor chasing the fastest published time what changed to get there, rather than treating the number alone as a reason to pick them.

What did Genlook just change, and does it matter?

Genlook, a named competitor already covered in Corlen vs Genlook vs Antla, released a new model on September 1, 2026. According to Genlook's own announcement, it generates a photorealistic result in under 10 seconds from a single uploaded photo, and extends reliable rendering to garment categories earlier versions struggled with, including footwear and accessories alongside dresses, outerwear, and swimwear. The upgrade applies to every store already on the platform at no added cost.

That is a real, current move in the category, and it matters most to a merchant already comparing Genlook against alternatives on more than one axis. Speed is one input into that decision. Category coverage, privacy handling, and where the tool runs beyond a single storefront are the others, and the full comparison lives in the linked post above rather than repeated here.

How fast is Corlen, and why not chase the fastest number alone?

Corlen states 10 to 15 seconds per try-on in balanced mode, the same figure on its in-store kiosk and its Shopify app. The developer API runs the same engine but returns the result as an asynchronous job, polled every few seconds until it finishes, rather than a fixed synchronous wait, which matters for a developer building around it more than the raw seconds figure does.

That number is not the fastest on the market today, and it is not trying to be. Corlen runs on Gemini 3 Pro Image specifically because it holds a shopper's face, skin tone, and the room around them steady across the range of ordinary phone photos real shoppers upload, a tradeoff already covered in the AI models behind virtual try-on. A merchant deciding between vendors should ask for the seconds figure, but should also ask what a vendor gave up, or gained, to land on that number.

See the actual wait for yourself: try Corlen on your own photo, no account needed for a first result.