About

About TIA

TIA checks whether the text inside an image meets WCAG contrast requirements. Most automated accessibility checkers read the page’s code, so they skip text that is part of a picture: banners, screenshots, social cards, and ads. TIA reads the picture.

What it does

Give TIA an image and a WCAG level, AA or AAA. It finds the text in the image, measures the contrast of each text region against its background, and gives each region a verdict: Pass, Fail, or Review. It works on text over photographs, gradients, and flat color. You can use it in the browser, from a script through the HTTP API, or from an AI assistant through the MCP server.

TIA is an automated aid. It is not legal advice, and a Pass is not a certification of conformance.

Who it is for

• Designers checking a banner or social card before it ships.
• Developers who want a contrast check on images in a build or a CMS.
• Content editors who publish images with text on them.
• Accessibility testers who need a first pass over many images.

Where TIA can be wrong

TIA is a beta and is still improving. Each item says which way the mistake goes, so you know what to check by hand.

What it is tested on

• Works on text over photographs, flat color, gradients, and smooth imagery. The hardest case is text that sits on busy photo detail and is close to the required contrast: check a Pass there by eye.
• Non-Latin scripts (for example Devanagari or Thai) are outside what we have tested. Text size is judged cautiously or not at all, so expect more Review results.

Where it can be wrong

• Clean interface screenshots of apps and web pages are a weak spot today. TIA sometimes marks readable text there as Fail, especially small or bold text on flat colors. Double-check those results by hand before you change a design.
• Large text: to avoid false passes, text is only judged against the lower large-text threshold (3:1) from roughly 31 to 34 px up (all-caps headings are recognized as large from about 33 px). Headings between 24 px and about 33 px will often get Review instead of Pass.
• Bold and heavy text can be wrongly marked Fail at about 1:1, even when its contrast is strong. For dark text on a light background this starts at about 32 px for bold (we have also seen it at 16 px), about 64 px for semibold and about 96 px for regular text; extra-bold or black text can fail at almost any size. Light text on a dark background fails sooner: we have seen semibold fail at 40 px. This mistake errs toward Fail. If a heading with obviously strong contrast fails at about 1:1, check it by hand. (Bold text can also wrongly pass: see the next item.)
• Bold text touching a shape of a similar color, such as a dot, can be wrongly marked Pass. If bold text touches another shape, check it by hand.
• Small or thin text on a busy background can be wrongly marked Pass if one letter sits on a patch close to its own color: our own tests do not check such letters. Check small text over busy backgrounds by hand.

What the verdicts mean

Review means we could not certify the text either way. It is never a hidden Pass: check it by hand. Fail means the measured contrast is below the WCAG requirement. Pass means it met the requirement within these limits. Reading a report explains the rest.

Privacy

• Your image: when you check an image, it goes from this site’s server to TIA’s analysis service, which checks it and sends back the report. TIA does not save your image or the report.
• Image URLs: when the MCP tool is given an image_url, TIA’s server downloads that image to check it. It only fetches public https addresses.
• Your IP address is used to apply the rate limit. It is held in memory for about a minute and is not stored.
• No accounts, no cookies for tracking, and no analytics. The hosting providers may keep standard request logs, such as the time and IP address of a request.
• The downloaded contrast-report.json is made in your browser from the report on screen.

Uploads can be up to about 4 MB. Do not upload images you are not allowed to share.

Sample credits

The three photograph samples are photographs released under CC0 1.0 (a public domain dedication) on Flickr, found through Openverse. CC0 asks for no attribution; it is given here as a courtesy. We added the text on them. The license is as declared by the uploader; we did not verify that the uploader holds the rights. The other three samples were drawn from scratch and contain no third-party image.

• Mountain photo on Flickr (CC0 1.0)
• Autumn photo on Flickr (CC0 1.0)
• Street photo on Flickr (CC0 1.0)

Who made it

TIA is built by Dohun Kim. It is a free beta.