Client preview
Virtual Try-On
Drop in a garment and see it worn either on one of our reference models or on a photo you upload. Every run is real GPU inference; the exact compute cost is shown per generation so you can judge the economics, not just the result.
What's running under the hood
An honest view of the pipeline and where the money goes.
ICCV 2025 try-on diffusion model
GPU inference, warm container
Human-parsing + OpenPose + DensePose
This client-facing frontend
Generation history (this gallery)
Input + output image storage
The diffusion model and preprocessing run on Modal (serverless GPUs, billed by the second). This page and its database run on Vercel. The backend stays on Modal; only the frontend moved here.
What a generation costs
GPU time × per-second rate. Cold starts add a one-off ~1–2 min on the first run after idle.
Existing model
~$0.0108
~14s on A100-40GB
Your own photo
~$0.0157
+ ~16s preprocessing on A10G
A100-40GB
$2.78/hr
Diffusion inference
A10G
$1.10/hr
Custom-photo preprocessing
Measured average over the last 3 generations: $0.0226 each.
Try it
Upload a clothing image, then pick a model or upload your own photo.
1 · Choose a model
2 · Your garment
3 · Options
Result
Result appears here
Previous generations
Stored in Postgres + Vercel Blob kept across deploys.
