One photograph.
Briefly seen.

The moment you select a photograph, the browser scales it down to a 1600-pixel JPEG, quality 0.88. The original on your device is never opened, copied, or sent. Only the resized version is uploaded.

If the upload fails for any reason — slow network, refresh, connection drop — nothing has been transferred and nothing persists.

The resized photograph is sent to the model that composes the reading and to the model that composes the projected portrait. That is the entire route. The bytes are not indexed, not classified against a permanent identity, not appended to any face database.

The composing servers run inside Google Cloud, inside our own private project, with no third-party access.

Within twenty-four hours every copy of the resized photograph is removed from our infrastructure by a scheduled deletion. The reading itself remains, attached privately to your code; the projected portrait remains; the bare upload is gone.

If you choose to publish a shareable page, the bare photograph is opt-in — never published without your explicit consent on the share screen.

Your photograph is never used to train a model — not ours, not anyone else's. There is no internal dataset of faces. The composing models we use are commercial APIs whose terms prohibit training on submitted images, and we have signed those terms in writing.

Your photograph is never published to a gallery, leaderboard, social proof grid, or any public surface unless you click the share toggle yourself.

If you would like every trace of the reading deleted — including the published share if you made one, including the stored Atlas, including the issued code — write to support@vaesyn.com. The deletion is permanent and processed the same business day.