SideNote ProAsk AI about scanned PDF pages
A scanned PDF holds pictures of pages rather than text, so a text model has nothing to read. SideNote Pro can draw the pages you choose as images on your PC and send them, with your question, to a model that can see them.
Before you start: a model that accepts images
Page images travel in your request as ordinary images, so the profile you send with needs a model, and an endpoint, that accept image input. SideNote Pro does not guess this from a model's name: a provider that cannot take images answers as it would for any image, usually with an error. Many small local models are text only.
A hosted provider may charge for page images under its own pricing, usually as image tokens; an image can cost more than the same page as text. The licence does not include AI usage, and a model run locally in Ollama or LM Studio has no API charge.
Read PDF pages as images, step by step
- Attach the PDFUse Attach documents in the composer, or drag the file onto the window. SideNote Pro first tries to read its text, on your PC.
- Continue when askedIf no page has readable text, SideNote Pro asks No readable text was found. Read PDF pages as images? Continue opens the page selector, and Cancel leaves your message as it was.
- Or start from the attachmentFor a PDF already attached as text, open its chip's More actions and choose Read pages as images…. A PDF with some pages without text says so on its chip, for example 2 of 4 pages have no text; they may be scanned or blank.
- Choose the pagesThe selector lists every page by its number, marks the ones without text, and previews the page in focus. Arrow keys, Home, End and Space choose pages from the keyboard.
- Attach themAttach N pages draws the chosen pages on your PC and adds each to your message as an image, shown as Page 3 · report.pdf. You can cancel while they are drawn.
- Ask, then press SendNothing is sent until you press Send. The page images go to your provider with your question, each labelled with its file and page number.
Five images per message, pages included
A message holds up to five images, and each PDF page takes one of them, alongside any screenshots or other images in the same message. The selector shows how many slots are free and does not let you choose more.
A PDF with more pages than free slots starts with nothing selected, so it is never cut to its first pages without you choosing them. When the pages without text fit, they start selected. For a longer document, ask about a few pages at a time, in follow-up messages.
Converting a text attachment replaces it
When you read pages of an attached PDF as images, the pages you choose replace that PDF's whole text attachment. The text of the pages you did not choose is no longer sent from it. If those pages matter too, ask about them before you convert, or choose them as images as well.
Converting reads the file again and goes ahead only if it is unchanged since you attached it. If you moved or edited the PDF in the meantime, SideNote Pro says so and keeps the text attachment. Cancel, or a failure, keeps it too.
What stays on your PC, and what is sent
Stays on your PC
The PDF itself. Pages are drawn on your PC by Windows' own PDF renderer, and a preview is drawn only for the page in focus. No text recognition runs in SideNote Pro. Page images are saved with the conversation like any other image, and protected with it when history protection is on.
Goes to your provider
The page images you attached, when you press Send, to the provider in the profile you send with. What the model reads from them is its own interpretation.
What to ask
Name the page and the output you want. A page image with no question tends to come back as a description of the page.
- Transcribe page 2 exactly, and mark any word you cannot read with [unclear].
- Explain the diagram on page 3: what each part is, and how they connect.
- Extract every date and obligation from these pages as a table, with the page each one comes from.
- Compare the signed page with the clause I pasted below, and list any differences.
A model's reading of a scanned page can be wrong, especially with handwriting, small print, stamps or a poor scan. Check anything that matters against the page itself.
Limits
Your provider's own limits on image size, image count and context still apply on top of these.
| Limit | Value |
|---|---|
| PDF file size | 20 MB |
| Pages in the PDF | Up to 500 |
| Images per message | 5, shared with screenshots and other images |
| Page size | Drawn at up to 300 dpi, then reduced to at most 2,400 pixels on the longer edge and 6 million pixels |
| Each page image | Up to 8 MB |
| Pages from one PDF together | Up to 20 MB |
| All images in one message | Up to 50 MB |
When it does not work
- The answer ignores the pages
- The model or endpoint may not accept images. Choose a model with image input in the composer, and check your provider's documentation.
- A page cannot be chosen
- All five image slots are in use. Remove a screenshot or another image, or ask about fewer pages.
- The PDF is refused
- A damaged, encrypted or oversized PDF stays refused, and is never retried another way.
- Conversion stops
- The file was moved or changed since you attached it, or you cancelled. The text attachment stays as it was.
- The selector closed by itself
- Starting or opening another conversation, or quitting, cancels it, and nothing it drew reaches either conversation.
Folders stay text only
Reading pages as images is for a PDF you attach to a message. Local folder context indexes text only, so a scanned PDF in an indexed folder contributes nothing to its answers, and SideNote Pro never sends folder files as images.
For a PDF with readable text, asking AI about documents covers attaching it and the prompts that work.