A virtual try-on API for a multi-brand retailer or marketplace has to answer a question single-brand buyers never face: can one model’s strengths and blind spots be applied uniformly across every brand and garment category in the portfolio, or does each brand need something different? OTB Group — the parent of Diesel, Jil Sander, Marni, and Maison Margiela — became the clearest public test case for this in 2026, when it announced on May 7 that it had put Google Cloud’s Vertex AI Virtual Try-On live for Diesel and Jil Sander across the US and Europe, with the other two brands planned to follow in subsequent months.
That deployment matters beyond its own headline because it makes the multi-brand constraint concrete: Vertex AI Virtual Try-On is one model, applied the same way whether the brand sells structured tailoring or streetwear. A tailoring house and a swimwear label do not stress a virtual try-on model the same way, and no single model in this report wins across every garment category. The practical decision for a multi-brand buyer is not just “which model produces the best image” but “does my portfolio need one model everywhere, or different models for different brands.”
TryOn API ranks first for exactly that reason: it routes 14 virtual try-on and generative-AI models across 7 providers through a single OpenAI-compatible endpoint, so a retail group can send an outfit-heavy brand to a multi-garment model, a swimwear brand around a model that restricts that category, and a basics brand to whichever option is cheapest — all from one integration. Google Cloud’s Vertex AI Virtual Try-On ranks second on the strength of being the only option here with a named, independently reported multi-brand production deployment, even though it is architecturally single-model. Fashn.ai ranks third as the fastest, most transparently priced path to production for a portfolio whose brands share a broadly similar aesthetic. Below we rank all six options on routing flexibility, verified pricing, multi-garment capability, and license terms.
Why one model rarely fits an entire brand portfolio
Virtual try-on models are not interchangeable commodities. FLUX VTO, launched by Black Forest Labs on May 28, 2026, is the only model in this comparison that composites up to four garments simultaneously with realistic layering — a real advantage for an outfit-driven brand, but its default moderation policy excludes swimwear and lingerie outright, which would immediately disqualify it for any swim or lingerie label sitting elsewhere in the same portfolio. Kling Kolors handles single garments reliably at a confirmed $0.07 per image but caps out at 768x1024 pixel output and struggles with garment-type switching. Fashn.ai’s dedicated e-commerce lineup is the most transparently priced but offers no per-brand model selection. A portfolio that picks any one of these as its sole provider is implicitly deciding that every brand it owns will live with that model’s specific weak spot.
What the OTB Group deployment actually shows — and does not
OTB Group’s rollout is useful evidence that a real multi-brand fashion house judged virtual try-on ready for production, not a pilot. It is not evidence that a single model serves every brand equally well: the announcement describes the tool as a premium clienteling feature — advisors sharing curated previews with selected customers — rather than a self-serve try-on available to every shopper across every brand’s storefront, and OTB rolled it out to two brands first, with two more still pending as of the announcement. That sequencing is consistent with the pattern this report keeps surfacing: even a well-resourced enterprise buyer is validating a single-model deployment brand by brand rather than assuming uniform fit across a four-brand portfolio on day one.
The license filter still applies at marketplace scale
Commercial license terms matter more, not less, once a marketplace has many sellers rather than one brand’s legal team reviewing a single vendor contract. Replicate’s IDM-VTON, at roughly $0.023 per run, is the cheapest model in this report by a wide margin, but its CC BY-NC-SA 4.0 license is explicit that commercial use is not permitted. A marketplace that let third-party sellers’ listings run through that model in production would be violating its terms at every seller, not just once. It is a legitimate tool for evaluating whether try-on technology is worth adopting before committing budget, and nothing more.
Where this is heading
The direction of the market favors flexibility over a single dominant model. Four credible foundation-model options now exist where fewer than three did a year earlier, each with different strengths — FLUX VTO on multi-garment compositing, Fashn.ai on pricing transparency, Kling Kolors on production maturity, Vertex AI on enterprise trust signals like SynthID watermarking. For a single-brand storefront, picking the best model today and re-evaluating annually is a reasonable strategy. For a portfolio spanning multiple brands, garment categories, and content policies, betting on one model’s roadmap is a bigger risk than it looks, which is the structural case for a routing layer that can absorb whichever model the market favors next without a re-integration at every brand.