AI features and what data they use

Automatic culling

Runs on the platform's own servers, using image measurements: sharpness, brightness and a small fingerprint of each frame to find near-duplicates.
No paid AI model is used, and no photo is sent to another company.
It works from the small thumbnail of each photo. The measurements are saved with the photo and deleted with it.
Your own verdict always wins over the machine's.

Off unless the studio switches it on for a gallery.
Before a guest takes a selfie, the page tells them in plain words that it is used once to find their photos and deleted right after.
The selfie is stripped of its camera data, used once to find the guest's photos, and deleted right after, normally within seconds. A selfie whose search never finished is deleted within an hour.
Faces are measured by a model server run by the platform, not by an outside AI company. The server keeps no images.
The face measurements of the gallery's photos stay with that gallery, and are deleted when its closing date passes, when selfie search is switched off, or when the gallery is deleted.
Nothing about a face is written to logs, and selfies never appear in a link.
People in the photos who never search are measured too, because their faces are in the photos. Studios should say so in their contract with the client.

AI help drafts replies and summaries on a few pages, and the assistant answers questions in the app. Both send text to Anthropic's Claude, or to OpenAI when Claude cannot answer and the platform has OpenAI set up.
What is sent: the question, the help pages it needs, and the studio's own records it looks up with the person's own access, such as an enquiry's name, occasion, dates, city and message, a project's events and crew, and, for people who can see money, amounts.
Clients' phone numbers and email addresses are taken out before anything is sent.
Neither one changes, saves or sends anything in the studio. A support request goes to the isodesk team only after the person reads it and presses send.
Conversations with the assistant are kept for the person who had them, who can delete them. When someone sends a support request from a conversation, the isodesk team can read that conversation, and the request says so before it is sent.
Phone numbers, card numbers, ID numbers and one-time codes a person types are taken out before the question is kept or sent.
When the help does not cover something, or someone says an answer did not help, the isodesk team sees that question, with those details taken out, so it can write the missing help.

Off until the studio switches it on.
It sends the studio's published page details and what the visitor writes to Anthropic's Claude, or to OpenAI as the backup.
If the visitor asks it to pass on an enquiry, the details they give go to the studio like the page's enquiry form.
Visitors are not signed in; a random browser cookie keeps their conversation, and conversations are deleted after 30 days.

Describe your event, and the wording on invitations, posters and thank-you cards, send what you type and the event's details (names, kind of event, dates, places) to Anthropic's Claude. Choosing the kind of event and ordering the suggested designs may use OpenAI.
You read every word before it goes on a card.
An AI picture is drawn from a prompt made of the design's colours and ornament only: never a name, a phone number, an address or a photo. It goes to Google's Gemini image model, or to OpenAI's when that is set up instead.
Each of these is off until the platform sets it up; the designs and cards work without it.

Photos are never sent to an outside AI company.
Your photos are not used to train AI models.
Nothing is sent to a language model unless someone uses AI help, the assistant, a studio's client chat or the AI on event cards.

Answers come only from the pages of this website.

Updated 08/11/2026 · Markdown version