A small service with limited data.

What we process

When you request a scene, SOILIE-3D processes the selected object labels, room settings, generation mode, seed, a random request identifier, and the generated files. To enforce anonymous quotas, the API derives a one-day identifier from the request’s source address using a secret-keyed hash. The application does not store or log the raw address.

The site uses local storage for theme preference, the current job identifier, and a short-lived management token. It uses session storage to decide when to show the optional loading transition. It does not use advertising or cross-site tracking cookies.

AI pilot and future human comparison

The Research page publishes AI-labelled responses, prompts, model identity, and frozen scene references. These records are retained as research evidence and are never counted as human participants. Private invitations and session credentials are excluded from the public exports.

There is no public comparison task, and human enrollment is closed. A human-participant study is planned. Before any human responses are collected, this notice and the study-specific consent materials will be updated. Participant codes should be used instead of legal names wherever the approved protocol permits it.

How information is used

Generation request data is used to validate relationships, queue and render scenes, show job progress, enforce service limits, reproduce outputs, investigate errors, and operate the public gallery. Structured operational logs contain job identifiers, processing stages, timing, response status, and stable error codes, but not complete request bodies, names, gallery passwords, or visitor addresses. AI-pilot responses are analyzed separately from any future human study. No human comparison responses are collected while enrollment is closed.

Retention and public gallery

Temporary generation files live under a dedicated storage prefix and are scheduled for deletion after seven days; job and idempotency records may remain for approximately eight days so expiry and failures can be reported. Choosing “Discard now” deletes the stored artifacts immediately and requests removal from the delivery cache, which can take a few minutes.

If you choose to publish, the displayed name (or “Anonymous”), object labels, room dimensions and fitting status, generation mode, seed, creation time, renders, animation, request, placement CSV, and manifest become publicly accessible and remain until removed. The four manuscript examples on the homepage are maintained separately and are never added automatically.

Publishing requires a removal password. SOILIE-3D stores only a salted PBKDF2-SHA256 verifier in a separate encrypted S3 bucket with public access blocked; it does not retain the plaintext password. Passwords cannot be recovered.

Infrastructure and external links

Amazon Web Services provides API delivery, Lambda processing, queues, database records, object storage, content delivery, and operational logs in Canada where configured. These services necessarily process technical request data under their own terms. Links to GitHub, the legacy Science of Imagination Lab page, and other research sites take you to services governed by their own privacy practices.

Your choices and contact

You may leave a scene temporary, discard it, or publish it. A public scene can be removed at any time from its gallery card by entering its password. If you lose the password, request a manual takedown and identify the scene URL. Local theme and job data can be cleared through browser storage controls.

For privacy questions, correction requests, or gallery takedowns, contact mike.cichonski@gmail.com. This notice may be updated as the research service changes; the effective date above will change with material revisions.