Photo-realistic try-on images
Diffusion-based try-on models generate a new image of the shopper wearing the chosen garment, including folds, drape and how it sits on their body, rather than overlaying a flat cut-out on a video feed.
A virtual try-on kiosk lets a shopper see themselves wearing a kurta, saree blouse or jacket without queueing for the trial room. Velozity builds AI try-on kiosks and APIs for fashion and jewellery retailers, using open diffusion-based try-on models: the shopper takes a photo, picks an item, and gets back a realistic image of themselves wearing it from the front and two angles. It connects to Retail OS or your own catalogue.
A shopper who would happily consider ten outfits tries three, because each one means a queue, undressing and redressing. Fewer tries means fewer chances to find something they will buy.
Stores carry a fraction of the catalogue in every size and colour. A shopper who wants the same design in maroon, which is only in the warehouse, has to imagine it. Most cannot, and leave.
Earlier AR mirrors pasted a flat garment over a video of the shopper. It looked like a sticker, ignored how fabric falls, and shoppers treated it as a toy rather than a way to decide.
Diffusion-based try-on models generate a new image of the shopper wearing the chosen garment, including folds, drape and how it sits on their body, rather than overlaying a flat cut-out on a video feed.
The kiosk can generate a front view and left and right three-quarter views of the same outfit, guided by reference poses, so shoppers see more than one flat angle before deciding.
Face recognition models extract the shopper's facial features so the generated image still looks like them, not a generic model. This is what makes people trust the result.
Run it as a standalone kiosk with a screen and camera, on a staff tablet on the shop floor, or as an API your website and app call. The same models and catalogue sit behind all three.
Each garment image is tied to its SKU, sizes and price. A shopper who likes the result can ask staff for it, send it to their phone over WhatsApp, or add it to a cart if you sell online.
Shoppers consent before their photo is taken, photos are deleted on a schedule you set, and the system can run on a GPU server inside your store or data centre so images never leave your premises.
We run the try-on models on 20–50 of your real products and a set of volunteer photos. You see honest results, including where it struggles, before committing.
We pick models whose licences allow commercial use, tune them on your product types, and size the GPU hardware for the number of shoppers you expect.
We design the screen flow: consent, photo, browse, wait, result, share. The wait is designed in, with the result sent to the shopper's phone so the kiosk is free for the next person.
One store runs it with staff trained and usage tracked. Once it works there, we roll out to more stores or open the API to your website.
Test try-on quality on your own products before you spend more.
A production kiosk for one flagship store.
Kiosks across stores plus try-on on your website or app.
Indicative ranges, in line with our computer vision projects. Kiosk hardware and GPU servers are extra; our on-prem AI server is one option if images must stay on your premises.
| Feature | Velozity | Typical alternatives |
|---|---|---|
| How the image is made | AI generates a new image of the shopper wearing the item | Garment overlay on live video |
| Fabric and drape | Folds and fit rendered in the image | Flat, sticker-like overlay |
| Speed | A short wait for each result on a GPU | Real time |
| Identity | Shopper's face preserved from their photo | Live video of the shopper |
| Where it runs | Kiosk, tablet or API, on-premise or cloud | Usually a dedicated mirror unit |
Our JewelCRM runs customer relationships for jewellery retailers, and our JewelCRM retail chain case study shows how we roll software out across stores. Retail OS handles billing and store operations. Virtual try-on is built by the same Chennai team of 50+ engineers, and it plugs into those systems so a try-on turns into a customer record and, ideally, a bill.
See Retail OSIt is a screen with a camera, usually in a store, where a shopper takes a photo, picks a product and sees an AI-generated image of themselves wearing it. Ours uses diffusion-based try-on models, which generate a realistic new image rather than overlaying a flat garment on video.
We recommend starting with a ₹3–5 Lakh proof of concept on your own products. A production kiosk for one store typically costs ₹15–30 Lakhs, and a multi-store rollout with a web API is ₹30–60 Lakhs. Hardware is extra.
No, and we say so up front. Generating a realistic image takes a short wait on a GPU, longer for multiple angles. The kiosk is designed around that wait, and results can be sent to the shopper's phone.
Upper-body garments such as kurtas, shirts and tops are the most reliable today. Sarees, lehengas and jewellery are harder and are tested product by product in the proof of concept, so you know what works before you pay for a full build.
We build on open diffusion try-on research such as IDM-VTON and IMAGDressing. Some public checkpoints are licensed for research only, so for commercial deployments we choose models and weights whose licences permit commercial use, and tune them on your products.
Shoppers consent before a photo is taken. Photos are deleted on a schedule you set, and the system can run entirely on a server in your store or data centre. We design to India's DPDP Act requirements for consent and deletion.
Yes. The same try-on engine can be exposed as an API so your website or app lets shoppers upload a photo and try products at home.
Send us 20 of your products and we will tell you honestly how well virtual try-on works on them, before you commit to a kiosk.
Call us: +91 80720 64524