Undress a photo in three taps — upload, style, download

The file never touches permanent storage. Here is every step between your tap and the output.

When you undress a photo through Undressia, the file uploads to an ephemeral container allocated just for your session. EXIF metadata — GPS, device model, timestamps — is stripped in the first processing step, before the image even reaches the diffusion model. The stripped original and the generated output both exist only in volatile memory.

The model segments the clothing boundary, reconstructs the underlying anatomy, and composites the result back onto the untouched background. The entire pass takes between six and nine seconds depending on resolution and clothing complexity. Multi-layer outfits (jacket over shirt over undershirt) are the slowest because each boundary adds segmentation cost.

Once you download or close the tab, both the upload and the output are purged. There is no gallery, no thumbnail cache, no admin-accessible archive. If you want the result again later, you process the photo again from scratch.

Resolution matters more than file size. A 3 MB JPEG from a modern phone camera will outperform a 12 MB upscaled image because the model needs real texture data, not interpolated pixels. If you have multiple versions of the same photo, choose the one closest to the original capture.

Lighting consistency helps too. The model reconstructs skin tones from the surrounding pixels, so even illumination across the subject gives it a better reference palette. Mixed lighting (half sun, half shadow cutting across the torso) is the most common source of colour banding in the output.