Skip to content

Which image formats can you upload to ChatGPT, and the 20 MB cap

What ChatGPT acceptsLast checked

Short answer

The common web formats work: PNG, JPEG, WEBP and GIF. Each image is capped at 20 MB, and there is a ceiling of about 100 images per conversation. Design and camera formats such as PSD, AI and RAW are not web images and need exporting first.

What to use

FormatVerdict
PNGBest for screenshots, diagrams and anything with text in it
JPEGBest for photographs
WEBPWorks, common on the web
GIFWorks, treated as a still
HEICApple's format. Convert to JPEG to be safe
PSD, AI, RAWNot web images. Export a flattened PNG or JPEG
SVGA vector format, and better handled by pasting the underlying code as text

As with document formats, OpenAI publishes no explicit list. What is documented is the 20 MB per image cap. The common formats work, and anything unusual is faster to convert than to test.

PNG for anything with text in it

Worth knowing if you upload screenshots.

JPEG compression works by discarding detail, and the detail it discards most visibly is sharp edges, which is exactly what letterforms are made of. A JPEG screenshot of small text gets fuzzy in a way a PNG does not, and fuzzy text is misread rather than flagged as unreadable.

PNG for screenshots, code, diagrams and documents. JPEG for photographs, where its compression is doing what it was designed for.

The limits

20 MB per image. Generous. A phone photo is a few megabytes and a screenshot far less. If you hit this, resizing solves it and loses nothing that mattered.

About 100 images per conversation, and PDF pages count toward the same ceiling. It is per conversation, so a new chat resets it.

When to send text instead

The more useful question than which format, and the answer is usually text.

Sending text is better when the content is text. It is exact. Nothing has to be interpreted, so nothing can be misread. It is searchable and quotable, and it costs a fraction of the length.

Sending an image is better when the layout is the point. A chart, a diagram, a form, a design, a screenshot of an error including its surrounding interface.

A misread screenshot fails silently

If ChatGPT misreads a number in a blurry screenshot, you get a confident answer built on the wrong figure and no indication anything went wrong. Sending the number as text removes that entire class of error. It is worth the extra ten seconds on anything you will act on.

Images uploaded directly are read. Images inside a document are not.

Every plan except Enterprise extracts the digital text from an uploaded document and discards its images. So a chart in a Word file or a diagram in a PDF is not available, and nothing says so.

If a figure inside a document matters, screenshot that page and upload it as an image alongside the document. That is the reliable way to get both.

The same rule is why a scanned PDF returns nothing: it is entirely images, so the extraction finds nothing to extract.

Common questions

What image formats can I upload?

The common web formats, PNG, JPEG, WEBP and GIF, are the ones to rely on. OpenAI publishes no explicit list, but images are capped at 20 MB each and the widely used formats work.

Can I upload a HEIC photo from my iPhone?

HEIC is Apple's format and less universally handled. Converting to JPEG first removes the question entirely, and most Mac and iPhone export flows will do it for you.

What about PSD, AI or RAW files?

Those are design and camera formats rather than web images, and they fall outside what is supported. Export a flattened PNG or JPEG of what you want looked at.

Is uploading a screenshot of text a good idea?

It works, but sending the text itself is better whenever you can. Text is exact, searchable and quotable, while a screenshot has to be interpreted and any misreading is silent.

Keep reading