Virtual Try-On HD

Upload a photo of yourself and a reference photo of any outfit — AI dresses you in it. HD tier: byte-perfect identity preservation outside the body region (face, hands, held objects, background all stay exactly the same), the new outfit is rendered inside via Klein 9B + the fal Klein VTO LoRA. ~10-15 min per output. The slow but seamless option.

Advertisement
Drop image here or click to upload
JPG, PNG, WebP · up to 25 MB
Drop image here or click to upload
JPG, PNG, WebP · up to 25 MB

Leave blank to let the reference image speak for itself.

Advertisement

How to Use Virtual Try-On HD

  1. Upload a photo of yourself (clear front-facing body shot works best)
  2. Upload a reference photo of the outfit
  3. Optionally describe the outfit in plain English
  4. Click Generate — output ready in ~10-15 min

Features

  • Two-image upload — your photo + the reference outfit
  • Identity stays locked: face, hands, held objects, background all byte-perfect
  • Klein 9B + fal Klein VTO LoRA + Klein Consistency LoRA stacked
  • Mask-based pipeline — no compositing seams
  • ~10-15 min per output (HD tier)
  • Pro plan only

Frequently Asked Questions

How is this different from the standard try-on?
Standard try-on (~60s) runs in edit-mode and regenerates the whole person — face, body, and outfit — then we composite the original face/hands back in. There can be subtle compositing seams. HD try-on (~10 min) runs in mask-mode: pixels outside the body silhouette are kept byte-for-byte from your original photo, so face / hands / held items / background never change. Slower but seamless.
Why does it take so long?
The mask-based pipeline runs Klein 9B inference at full resolution with the VTO LoRA + Consistency LoRA stacked. On the 2080Ti this takes ~10-15 min. We could make it faster with a stronger GPU; for now this is the trade-off for HD quality.
Will my face change?
No. The mask covers only the body silhouette. Face, hair, neck, hands are preserved as original pixels. Only the visible outfit area gets repainted.
What outfits work best?
Anything the LoRA can read from a reference image — dresses, jackets, T-shirts, jeans, suits, full looks. For complex multi-piece outfits, upload one reference at a time.