ChatGPT image upload limits: 20 MB, and why PDF pages count

The size limit is not usually your problem

20 MB per image. A phone photo is a few megabytes and a screenshot is far less, so unless you are uploading something from a professional camera, size is not what stopped you.

If you did hit it, resizing solves it instantly and costs you nothing that matters. ChatGPT does not need a 40 megapixel original to read a document or describe a scene.

The limit people actually hit is a count

The message reads roughly: you can add at most 100 images, including PDF pages, to a chat.

Two things in that sentence are worth pulling apart.

It is per conversation. Not per message, and not per day. Starting a new chat gives you a fresh allowance immediately, which is the simplest fix by a distance.

PDF pages count as images. This is the part that surprises people. Upload one PDF of 120 pages and you can trip an image ceiling having uploaded a single file. From your side you sent one document. From the counter's side you sent 120 images.

Reducing what you send

Crop before uploading. If your question is about one panel of a screenshot, send that panel. It counts the same as the full screen and gets a better answer, because there is less to interpret.

Combine screenshots. Several screenshots stitched into one image counts as one. Any image editor does this, and for a sequence of steps it often reads better anyway.

Convert PDFs to text. If the document has real text in it, exporting to text removes the page count from the equation entirely, because text is not images. This is the right answer for a long document you want read rather than looked at.

Send fewer pages. Extract the pages your question concerns into their own PDF.

Images inside documents are a separate thing

Worth understanding because it explains a different failure that looks similar.

Except on Enterprise, ChatGPT extracts the digital text from an uploaded document and discards the images. So a chart in a report, a diagram in a slide deck, or a screenshot embedded in a Word file is simply not there as far as the conversation is concerned.

It will not mention this. It answers around the gap, which is worse than an error, because the answer looks complete.

If it is a document rather than pictures

Most people who arrive at an image limit are actually trying to get a document read, and images were just how it happened to be stored.

If that is you, the image path is the hard way round. Getting real text out of the document, through OCR if it is a scan, and sending that instead sidesteps every image limit on this page, produces better answers, and is not capped at 100 anything.