Digital Dragons Inc

“If you wish to conquer the world, you best have dragons.”

— George R.R. Martin

Blind Creator Pre-Flight — Privacy Policy

Effective November 3, 2026. This policy describes Pre-Flight exactly as the current build works.

Blind Creator Pre-Flight is a pre-publish check for blind and DeafBlind creators, made by Digital Dragons Inc. Before a photo or short video is posted, Pre-Flight checks it and reports every result in plain text. This policy explains what data Pre-Flight processes, where that data goes, and how long anything is kept.

What this build is

The current build is a demonstration build. It works two ways: on example photos and videos — called takes — that are already stored on the Pre-Flight server, and on a photo of your own that you choose on the page and send to be checked.

In this build:

  • You can check one photo of your own at a time. Your browser converts the photo you choose to PNG and sends it to the Pre-Flight server, which keeps a working copy while you check, repair, and share it. That working copy is deleted within two hours. There is no video upload.
  • There are no accounts, no sign-in, and no passwords.
  • Pre-Flight never publishes anything. It does not connect to Instagram or any other social platform, and it cannot post on your behalf.
  • The Library Sweep in this build uses a seeded, simulated mini-library. It does not read your photo library.

If a future build widens what creators can check, this policy will be updated and republished before that build is released.

What data is processed

When you run a check, Pre-Flight processes:

  • A photo of your own that you chose, converted to PNG in your browser and sent to the server, or the take you chose: its image, or for a video take the still frames extracted from it when the example media was made. The running server does not decode video.
  • An own photo has no recorded facts and no stated intent. The checks are measured and judged from the photo itself, and where its subject cannot be located, framing and crop safety report uncertain and say why, rather than falling back to a record.
  • Recorded facts supplied with the example take, including its title, kind, the creator’s stated intent, subject and sign positions, and — for a video — its duration, crop window, banner text, whether it is silent, and relevant timings.
  • The choices you make while you work: which take to check, which repair to request, and whether a repair should be based on an earlier repaired version.
  • Whether you have acknowledged a private-information warning and approved the alt text. The readiness check receives those as yes-or-no values, not the edited text itself.

The alt text and optional public description are edited in your browser’s current page. The Pre-Flight code does not put those edits in cookies or browser storage. When you approve or edit them, the approval state is kept for that working session in the page.

The application code does not ask for or create a store of your name, account email address, location, contacts, photos from your device, or a device identifier. If you email us for support, that separate communication is covered below.

The Meta Model API’s role

Some checks are measured on the Pre-Flight server itself, from the take’s own pixels. Other checks use models provided by Meta through the Meta Model API. The Pre-Flight server — never your browser — sends data to Meta, and only when live model calls are switched on:

  • SAM 3.1 receives the take’s image — for a video, a still frame — with one concept at a time, such as “person” or “sign,” so framing and crop safety can be measured from the boxes it returns.
  • Muse Spark 1.3 receives the take’s image together with the creator’s stated intent. It judges the text in the image, the first impression, and whether anything private is visible. When SAM 3.1 finds no subject match, Muse Spark can also estimate the subject and sign positions; the result identifies that source as judgement, not measurement.
  • For a video’s described-audio draft, Muse Spark receives text only — the take’s recorded facts, and its transcript if one is available — not the image.
  • Muse Voice Transcribe is the transcription route for a video’s separate speech audio file. No example take in this build ships such a file, and a silent take sends nothing: there is no audio track to transcribe, and Pre-Flight says so instead of sending silence to the model.

No generative image model edits the media. Muse Image is not used in this build. Repairs are measurable operations the app performs itself, such as pulling the frame back or choosing another cover frame.

Pre-Flight’s backend is hosted on Fly.io, which operates the server that carries out this processing.

The contributor training tier is not used

Pre-Flight uses the Meta Model API on standard API data terms only. It does not use the discounted “contributor” tier, which permits training on prompts and completions. That is a standing product decision: the contributor tier is never used for creators’ media. The production code calls Muse Spark 1.3, not a contributor variant.

What is kept, and for how long

  • The example takes and their recorded facts are stored as part of the demonstration service so you can choose and check them.
  • While the backend runs, it keeps working data in memory to avoid repeating the same processing: media buffers, measurements, check results, transcripts when one exists, and described-audio drafts. The code sets no separate cache expiry; this in-memory data ends when the backend process restarts.
  • A working copy of a photo of your own is kept on the server while you check, repair, and share it. Working sessions older than two hours are swept and deleted; a sweep runs whenever a new photo arrives. Repairs to your own photo are made inside that same working copy, so they share its two-hour lifetime.
  • If you ask for a repair, the repaired copy of the example take and its updated recorded facts are written to the server’s repaired-media folder so the checks can run on that version. The original take is not changed. This build has no automatic deletion period and no visitor-facing delete control for repaired copies of example takes.
  • The proof copy, alt-text draft, public description, and described-audio draft are generated for display in your browser. Downloads are created by your browser as plain-text files. Pre-Flight does not post, attach, or share them for you.
  • The application code sets no cookies and uses no local or session storage for tracking.

Sharing, advertising, and analytics

Pre-Flight does not sell data. Its application code carries no advertising and calls no analytics or tracking service. Apart from the Meta Model API processing and Fly.io hosting described above, processing in the application stays between your browser and the Pre-Flight backend.

That statement covers the application code. It does not describe any logs or retention practices operated independently by an internet, hosting, model, or email provider.

Your choices and support email

Because this build has no accounts, there is no personal account data in Pre-Flight to view or correct. A working copy of a photo of your own deletes itself within two hours, as described above. If you have a privacy question about Pre-Flight — including about a future build that handles creators’ media more widely — contact us and we will answer in plain language.

If you email support, you provide your email address and whatever you include in the message. We use that information to read and answer your message. Please do not send passwords, private media, or other sensitive material unless we ask for a specific item and explain why.

Contact

Digital Dragons Inc shares knowledge and products that help people with disabilities succeed in the digital world.

Visit the main Digital Dragons Inc website