Your data and AI

What stays in your browser, what is sent when you ask for AI, and what a server operator can retain.

Ordinary editing and imports

Projects, imported images, videos, songs and generated results are stored in this browser's storage. Importing a file does not by itself upload that file to an AI provider. Output windows on this computer use those local assets. Exporting creates a file that you control; see Projects and backups.

When you ask AI to generate

The studio sends the requested prompt, relevant surface information and any applicable starting image through the prism server to the selected service. Anthropic handles procedural animation, shader and adaptation requests; OpenAI handles textures; Google's Gemini API handles video clips. A video request that uses the current picture as a starting frame sends that image too. Generated results are returned to the studio and saved locally. The server temporarily holds long-running job results for collection: completed jobs are retained for up to 30 minutes, subject to an overall memory limit; an older result may be removed earlier. Running jobs time out after 20 minutes.

The server holds provider credentials; they are not stored in project exports or sent to the phone. Read the selected service's account terms for its own retention and use of submitted data; prism's local-storage behaviour does not control a provider's retention. Only use content you intend to submit in a requested generation.

See AI animations for controls and Server setup for availability and limits. AI integrations are treated as validated for this release. Screenshots made with test replies remain labelled as illustrations.

A phone scan uses the server

The phone captures the projected patterns and prepares the scan results. Scan blobs and the resulting room photograph travel through the paired server to the studio. The normal server scan store has a two-hour expiry; rooms expire after twelve hours without activity. A downloaded scan saved in your project remains in your browser until you remove it.

If the operator configures SCAN_BLOB_DIR, the normal expiring store uses disk instead of memory. DIAG_DIR is different: it enables additional diagnostic logs and scan archives, with no automatic retention schedule. It is off by default. The operator must control access and remove those archives when no longer needed.

Copy diagnostics

The pre-show panel copies a small report: release, browser, connection states, output resolution and flash-limiter status, save state, audio state, asset counts and storage estimates. It excludes project and file names, project IDs, room URLs and tokens, provider keys, prompts, pictures and raw error text. Nothing is sent automatically; review the clipboard before sharing it.

Removing your data

Export what you need before deleting projects, media or browser site data. Deleting local data does not remove a backup you shared, an operator's diagnostic archive or a provider's records. Ask the server operator about server-held archives. This page describes prism's behaviour, rather than promising a third party's policy.

Reviewed for prism 2026.09.29.1 · 29 September 2026. Release notes.