Help

1. Getting started

1. Upload a book. Workspace → Upload, or drag an EPUB onto the page. Up to 250 MB.
2. Wait for the audit. It runs automatically and takes about a minute for a typical book. It checks the file against EPUB Accessibility 1.1, WCAG 2.2 AA and EPUBCheck.
3. Review the findings. The project page lists every issue, grouped by what it is and where it is.
4. Run the fix. Deterministic repairs — language attributes, landmarks, heading structure, metadata — are applied without asking. Alt text is generated for images that need it.
5. Check the alt text. This is the part that needs you. See the alt-text section below.
6. Export. You get the remediated EPUB plus a conformance report.

Nothing is charged until a fix runs, and the cost is quoted before it does.

2. What you can upload

EPUB only, up to 250 MB. EPUB 2 files are upgraded to EPUB 3 as part of remediation, which is charged separately per page.

PDF is not supported yet. It is in active development and the pipeline for it exists, but it is not part of the service today. Uploading a PDF will not work.

A file larger than 250 MB is rejected with a clear message rather than a timeout. If you have a book that genuinely exceeds it, email us — the limit is a transport constraint, not a policy.

3. The compliance score, and what it does not mean

Each book gets a percentage. It is the proportion of detected issues that are resolved, and it is useful for tracking progress on one book over time.

It is not a conformance certificate. A 99% score means one row is still open, not "nearly compliant" — and a single open image can be the one that matters. Read the report, not just the number.

We also do not claim the output satisfies any particular legal obligation. Conformance is a property of the file as you publish it, and you are the one publishing it.

4. Alt text

Images that need a description are sent to Azure OpenAI, a Microsoft service, along with a short instruction and up to 800 characters of the text immediately around the image on its page. That context is what lets the model describe what a figure is actually doing rather than merely what it looks like. The rest of your manuscript is not sent, and nothing you upload is used to train any model. The privacy policy has the full detail, including what Microsoft retains under abuse monitoring.

Generated alt text is a draft. Review it. A model is at its least reliable on an image whose meaning depends on surrounding context — a diagram that makes a point, a photograph used ironically, a chart whose message is the trend rather than the values. Every description is editable before export, and you can ask for a regeneration with a hint if the first attempt missed the point.

Where a book already has human-written alt text, we keep it. We do not overwrite a description someone wrote on purpose.

Two cases we flag rather than fix, because fixing them would destroy information:

• Alt text that repeats the caption. A screen-reader user hears the same sentence twice. Reported so you can replace it with a description.
• Alt text that is a credit line, such as "Courtesy of the National Archive". That is attribution, not description — and often the only place the credit appears in the book. We never overwrite it; move it to the rights metadata first, then describe the image.

Processing images on hardware we own, with nothing sent to any third party, is on our roadmap but is not something we offer today. If it is a requirement for you, tell us before uploading anything.

5. The Metadata tab

Open a book and there is a Metadata tab beside Source Editor and Image Manager. It edits the OPF directly: title, creator, publisher, language, identifier, date, rights, source and description.

Language and identifier matter for accessibility conformance, and rights is where an image credit belongs if you are moving one out of an alt attribute.

Edits are surgical — only the fields you change are touched, and the manifest, spine and every other part of the OPF are left byte-identical. `dcterms:modified` is updated only when something actually changed.

6. Credits and billing

The service is pay-as-you-go against a credit balance. New accounts get trial credit on first sign-in, enough for roughly one full-length book.

Cost is calculated per page and per image needing generated alt text, and is quoted before a job runs. Cover, half-title, title-page, decorative and duplicate images are not charged. Current rates are on the home page and in section 7 of the Terms, which reads them from the billing configuration itself so they cannot drift.

Credit does not expire. Unused credit bought in the last 30 days is refundable in full. If a job returns an empty or unusable file, or fails because of a fault on our side, email the project ID and we will return the full job cost.

Trial credit is limited to a fixed number of accounts. Past that, signups still work — you get an account and can add credit yourself, you just do not get the free grant.

7. When something goes wrong

The upload failed. If the file is over 250 MB you will be told so directly. Otherwise, a very large book can take a couple of minutes to transfer before anything appears — give it that long before retrying, because a retry starts the transfer again from zero.

Alt text generation did not start. If the fix is refused outright, the message says why and nothing is charged. If it starts and produces nothing, tell us the project ID rather than retrying — a repeat attempt on the same book behaves the same way, and we can see from the job what stopped it.

The page went blank or showed an error. There is a reference code on the error screen. Send it with your message and we can find the exact failure rather than guessing.

A fix ran but an issue is still open. That is often correct — some findings are reported rather than auto-fixed on purpose, because the automatic repair would lose information. The credit-line case above is the clearest example.

Something looks wrong in the output. Do not publish it. Email the project ID; we would rather look at it than have a file go out with our name on the workflow.

8. Your files, in short

• Your content is never used to train any model.
• Your books stay until you delete them. Nothing expires on a timer and nothing is removed on your behalf. Delete a single book from its project page, or everything from account settings.
• Export bundles in our archive are deleted automatically after 90 days.
• One third-party AI service is involved, only for image descriptions, and it is named in the privacy policy. It receives the image and up to 800 characters of the text around it — never the rest of your book.

The full detail, including what Microsoft retains under abuse monitoring, is in the Privacy Policy.

9. Getting a person

The Help button in the bottom corner of the app reaches us directly. Pick what it is about — something broken, billing, a question, or an idea — write the message, and it arrives with the page you were on attached. You get a reference you can quote. Or email connect@holographpressworks.com.

Sent from a book’s own screen, it arrives already attached to that book. From anywhere else, name the book or paste its project ID — it saves a round trip and lets us look at the actual job rather than reason about it.

We reply by email, usually within a working day. HoloRemedi is in beta: things change, and some of them change because someone wrote in. Bug reports are genuinely welcome.