Yes, I believe there is potential on this topic, and that we should address it.

However, I do not believe it to be very simple. Primarily because high-contrast means different things in different OSes and UAs, to which authors should react differently:

• First axis
  • Contrast-boost via image processing at the pixel level (possibly done by the OS with no understanding of the content)
  • Content-aware contrast-boost, which may pick different (contrasty) colors for different types of content, treats images differently from text, could affect which font or font-weight is used (necessarily done by the UA)
• Second axis: Is the contrast boost coupled with a color inversion (which can be done at the pixel level, or content aware to avoid inverting images), or with enforcing dark-on-light or light-on-dark (which assumes an understanding of what's content and what's background).

So, do we need 2 (or 12) completely separate queries to express all the various things that exist? 1 query with a bunch of values? Expose some higher level semantics? How?

I've heard ideas on this topic, and have a few of my own, but I don't think we have firm conclusions yet.

Absolutely. Having very detailed indication of what kind of high-contrast the OS/UA are actually using will be tricky. Was only thinking as a first stage having at least an on/off indicator, but agree that it's possibly of little practical use to devs beyond a basic "pick strong contrasting foreground/background color combinations" (which would still be strongly contrasted in situations where the color inversion is done at pixel/OS level, and would be replaced by OS' enforced colors in situations like Windows High Contrast).

Not sure even that would help. For instance, OS X (at the system level) has two settings that work on boosting contrast: one that works at the pixel level, and one that is more content aware. The one that's content aware also adds (in the native UI at least) borders around things. If the high contrast pair of foreground/background colors that a web author picks accidentally makes the background match the border, you may accidentally be making things worse. Also, I don't think it is useful for an author to react to content having been put in high contrast mode by putting content in high contrast mode, since it's already been done. Either the author needs to do it because there's a request to do so but the UA hasn't done anything itself, or because the author knows that a particular way to boost contrast gives undesirable side effects with their content, and want to remedy that problem, in which case they need to know what adjustment has been made.

Here are a few things we may want:

1. have colors been forcibly inverted at the pixel level (if yes, authors should do things like double-inverting images and removing shadows) (see https://drafts.csswg.org/mediaqueries-5/#inverted)
2. has contrast been forcibly boosted at the pixel level without inversion (if yes, authors should do what???)
3. Is the UA enforcing a content-aware color scheme (light-on-dark, dark-on-light...), or content-aware contrast-boosting measures other than the color scheme (think solid borders, sans-serif fonts, different styles of native components or focus hightlights)
4. Is the UA doing nothing particular, but suggesting (based on a user preference) that the author does the sort of things described in 3
5. Do we want to give authors a property that lets them disable UA-provided content-aware contrast boosts on a particular element, so that they can provide it themselves in a way that is more appropriate for their content? (like https://msdn.microsoft.com/en-us/library/windows/apps/hh441137.aspx)
6. Does this interact with https://drafts.csswg.org/mediaqueries-5/#light-level, and if yes, how?

@grorg I'm curious to have your/Apple's thoughts on this.

If the high contrast pair of foreground/background colors that a web author picks accidentally makes the background match the border, you may accidentally be making things worse.

not necessarily worse, provided that the foreground/background the author picked is high-contrast. the forced border would not show, but as that border is only there to ensure contrast (and not to make things fatter/bolder), the end result would still be a strong contrast, arguably.

but yes i can see how all this is far more nuanced than a simple true/false statement would allow for (which is likely why this hasn't been standardised until now).

I understand that this is supposed to be subsumed by https://drafts.csswg.org/mediaqueries-5/#light-level

However, I wonder if "high contrast" would be better as the primary concept, with light levels mapped onto the various high contrast preferences (dim -> light-on-dark, washed -> dark-on-light)?

I think that would better match the way people would tend to think about these options.

I understand that this is supposed to be subsumed by https://drafts.csswg.org/mediaqueries-5/#light-level

Maybe, maybe not. These things overlap for sure, but exactly how to sort that out is still unclear.

@frivoal

Maybe, maybe not. These things overlap for sure, but exactly how to sort that out is still unclear.

Good point. I chatted offline with @minorninth about this.

Some slightly scattered thoughts:

• light-level defines more of a "why" than a "what", and I think page authors would be more interested in "what".
• @minorninth pointed out that there are at least three axes as to "what":
  • Inverted colors: on/off
  • Grayscale: on/off
    • Plus more subtle colour modifications like "warm light" or "daltonized", maybe?
  • Increased contrast: none vs maximum (slider)
• And further, each of these may be:
  • controlled by the OS: the page needs to react to the "pixel level" change, or
  • expressed as a user preference: the page needs to apply the preference (e.g. "I am a hacker and I want light on dark theme for everything all the time so I can pretend to be in the matrix")
• User preferences may ideally be browser-wide or (like text zoom) site specific and frequently changed.
• It would also be nice to be able to configure the relationship between light level and the above factors, with some good defaults.

@alice: Thanks, a bunch of good points.

And further, each of these may be:

• controlled by the OS (pixel level)
• expressed as a user preference

I think there's one more:

• Controlled by the UA, taking the markup/semantics into account

I would like to suggest a prefers-high-contrast query, set by the UA either with a dedicated preference or by inheriting the OS accessibility settings (when relevant).

My main use case would be site designs that are expected to use a brand color which has bad contrast (when contrasted with a white background or white text). Switching the brand color with a more accessible one would be more acceptable to site owners if done for users who specifically requested it.

Currently, websites with this use case either don’t address the issue, or offer in-site controls like on sncf.com:

Options in this example are:

Contrast: default | stronger | inverted
Animations: enabled | disabled

@fvsch This issue is indeed about creating something like this, but I am not quite sure how your proposal answers the questions/concerns raised above it. The other prefers-* media queries are dealing with simpler topics, where there's just one axis, but this has multiple concepts overlapping, at it still isn't clean to me how to expose all the relevant parts in a way that is both sufficiently detail to be able to do the right thing, and sufficiently well designed that it is easy to do the right thing.

agree with @frivoal here. prefers-high-contrast indicates the user's wish. the discussion here is about detecting if the environment/OS is already set to deliver a different presentation to the user (and to then possibly take that into account when defining your own styles to take THAT already into account).

i could see potential for a prefers-high-contrast, separate from this discussion here. though then there's a (small?) risk that a browser would report that the user prefers high contrast, and a site changes its styles accordingly, but that the user also has their OS already doing some adaptation for them - leading to a doubling down on contrast potentially too far the other way? (e.g. if the author used the prefers-high-contrast as an indication to load in much more contrasty photos/images, which would then be made even more contrasty by the OS)

I am not quite sure how your proposal answers the questions/concerns raised above it

I’m not answering them and I know this is not an easy topic. :) I just felt this thread lacked input from authors about common design requirements and the client-side workarounds we use today (accessibility preferences that users have to find and enable on every site).

leading to a doubling down on contrast potentially too far the other way?

If a site switches to e.g. black text on white background on prefers-high-contrast, or shows black-on-white diagrams instead of diagrams with more subtle colors, I don’t think the OS contrast rendering pass would make it worse.

It might be an issue if authors use this preference hint to add body { filter: contrast(2); } or something similarly indiscriminate.

Chiming in with a real-life use case from a recent design project.

If the user has set increase contrast to “checked” in their host OS:

I would like to switch the orange in my designs to a11y orange in order to pass WCAG requirements.

It would change this:

To this:

As an author I would like to do it with the simplest code possible:

.button {
   background: #F96830;
}

@media (prefers-high-contrast) {
    .button {
        background: #D03F06;
    }
}

This thread might be duped out to either the original #443 or the more recent syntax discussion in issue #2943.

@Wolfr This actually seems like an anti-pattern to me.

"High contrast" is distinct from WCAG AA compliant contrast, and is typically used by users with unusual vision conditions (often, but not always, conditions which qualify as "legally blind" where such a category exists). It will often look something like this:

(Image from a blog post by The Paciello Group on -ms-high-contrast)

In, uh, contrast, WCAG AA (and preferably AAA where possible) compliant contrast should always be enabled for all users.

In, uh, contrast, WCAG AA (and preferably AAA where possible) compliant contrast should always be enabled for all users.

Yeah but the reality is that this rarely happens, the very Github comment button I am looking at does not pass WCAG AA.

Designers are really looking at WCAG contrast ratio numbers together with the visual end result to try and make an interface that is both accessible and aesthetically pleasing.

Here's an example from AirBnB where the colors are marked with “a11y”

My initial idea was that I wanted to be able to target users who have enabled a preference for high contrast.

What I found is that the thinking around this is that the contrast level should be specified in the media query.

My proposal is that it should be as simple as possible for the developer, otherwise they won't do it. Who has time to understand what 0.1 or 0.5 and 0.8 means in relationship to all the various factors at play? A singular switch seems like the most dev-friendly thing to do which in turn will lead to actual usage.

@alice Looking at your comment, it seems my proposal for syntax is confusing because of the existence of MS's full-on high contrast mode.

I wonder whether MS's full-on high contrast mode (almost like a Windows “theme” in my memory”) act s on every website as a sort of system override or does it only change the Windows UI and doesn't touch websites? I don't have a Windows device in hand to test.

Your older comments [in this thread] indicate that you have a deep understanding of the problem space. As a web developer I was mostly looking for more attention to this issue when I posted my initial comments.

The WCAG guidelines pose contrast fixes as one of the most important things one can do for a11y but in reality if you try to design with those AA/AAA passes in mind you will end up with something less than visually pleasing.

There's also often (inaccessible) brand guidelines that make it impossible to get your design approved if you would change the colors to follow AA/AAA logic.

So as a designer you are stuck: you want to do the right thing, but there is no way out.
But if you could implement some color exceptions with media queries that is a very clear solution.

@patrickhlauke can you advise if this issue is still relevant so APA should track, or if it's addressed somehow somewhere?

@patrickhlauke re-ping on question - is this issue is still relevant so APA should track, or if it's addressed somehow somewhere?

@michael-n-cooper sorry, not spent any time here or at/with APA, so no idea...but seems that @cookiecrook pointed to #2943 and #443 (the latter of which then points to #1709)

maybe a @w3c/csswg member can check/clarify if this thread is still needed / if the other locations should be tracked by APA instead

Thanks for the pings @michael-n-cooper.

I'm closing this issue because the original feature request has been fulfilled: a prefers-contrast media feature exists in the Level 5 spec, with options low, high, and no-preference. It is defined in a way that is orthogonal to color mode preference, ambient light level, and to whether forced color overrides are being applied.

Any ongoing debates about the details should be brought up in separate issues (e.g. #2943 about whether there should be a distinction between increased and high).

Pinging @frivoal, who is listed as the assignee, in case this needs extra tagging for a future Disposition of Comments.