@Bob Tregilus Only that "my best" has actually led to unimaginable extremes.

They say an image is worth a thousand words. I've once described one image in over 10,000 words. Over 60,000 characters. The post is so long that, I think, Misskey and its various forks have rejected it, as have Pleroma and Akkoma. It took me two full days, morning to evening, to describe that one image, in-world research included.

And I actually had to limit myself. For once, I did not give in-depth descriptions of the images within that image, especially not beyond what's actually visible in these images. That's because I've discovered that if I were to do that, I'd have to describe dozens of images in one particular image (in my image) and potentially over a hundred images in these, even though they're so small that they're technically invisible. It would have taken me months to write all that. And it would have been futile anyway. My character limit is over 16 million, but Mastodon rejects posts over 100,000 characters, and in the few places that do accept posts with millions of characters, next to nobody cares about image descriptions.

I haven't posted a new in-world image in over half a year. I've been working on-and-off on the descriptions for a series of rather simple avatar portraits since last autumn.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Bob Tregilus Of course, this means that the more obscure the content of your image is, the more in-depth you will have to go. At worst, there's nothing in your image of which non-sighted people know what it looks like unless you describe it. Simply mentioning that it's there is not sufficient.

My own original images aren't even photographs, nor are they pieces of art that represent real life. They're renderings from 3-D virtual worlds, very obscure 3-D virtual worlds even. Nobody knows what anything in these world looks like unless they can see it in my images. At the same time, however, chances are that they become so curious about these virtual worlds that they also become curious about everything in the image, not just what matters within the context of the post. That is, sometimes the image itself as a whole is the context. Either way, this means I can't just focus on certain elements in the image in my descriptions. I have to describe everything.

So I've gotten to a point at which even filling the alt-text character limit forced by Mastodon, Misskey and their respective forks (they cut longer alt-texts off at the 1,500-character mark) doesn't cut it. All my original images have two descriptions now. In addition to the one in the alt-text that's very limited, there is another one in the post that's more or less fully detailed, that contains transcripts of all text within the borders of the image, and that also comes with all explanations that I deem necessary. Since I don't have a character limit to worry about (the limit is defined by the database field rather than a hard-coded or configurable number), this description is likely to grow well over a hundred times longer than typical alt-text.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Michael Hanscom Out of curiosity, and to be on the safe side and improve my own image descriptions further: What makes alt-text bad in your opinion? Only when it's rubbish that doesn't have anything to do with the image?

Or is it bad when it isn't accurate enough? When it isn't detailed enough/when certain details that you think should be described are missing? When it's too detailed? When it describes the wrong details?

Is it bad when elements in the image are described the wrong way? When (for any definition of person) a person's skin colour is mentioned rather their race or vice versa, whichever you think is correct? When a person's gender is mentioned although whoever described the image can't know it with absolute certainty? When a shade of a colour is mentioned by name rather than being described?

Is it bad when it doesn't explain what you don't understand in such a way that you understand the image without having to look anything up yourself? Or is it bad when it does explain the image because explanations do not belong into the alt-text (I'm being serious, they actually don't)?

Is it bad when, while getting everything else right, it doesn't contain verbatim transcripts of any and all text in the image? What if there are transcripts in the post text body instead? Is it bad when text is transcribed which you think should not be transcribed, whichever that may be? Is it bad when text in a foreign language is transcribed verbatim in that language and then translated as literally as possible instead of translating it right away?

Or do you have other criteria that I don't know about and I haven't thought of yet?

When is alt-text so bad that it justifies both mocking and writing a replacement?

#Long #LongPost #CWLong #CWLongPost #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Justin Derrick The question, however, is: What is "high-quality"? How is it defined?

Would the bot go by the definition valid for commercial/scientific/technological websites and blogs, i.e. ideally no more than 125 characters, and only a short and concise visual description with no further information?

Or would the bot go by Mastodon's culture and Mastodon's standards, i.e. the longer and more detailed, the better, any and all extra information is welcome in alt-text (because it doesn't fit into the toot), and the limit is 1,500 characters?

That is, if it were for me, the bot would go look both for alt-texts and for image descriptions in the post text body and judge both. Because I do both at the same time for my original images. An extremely detailed long image description in the post itself (character limit for post and alt-texts combined here: over 16 million) that also comes with all necessary explanations and transcripts of all text in the image, plus an alt-text that's as detailed as 1,500 characters (minus notification about the long description in the post) allow, but with no explanations, and I usually have to leave out text transcripts as well because they're too many.

You may say the alt-text is superfluous if it's just a much shorter version of the long description. But as long as the Mastodon HOA demands there be an alt-text to every image, no matter what (especially seeing as I always hide my image posts behind summaries/content warnings, so you can't see right of the bat that there's a long image description in the post), I add alt-texts to my original images.

I'm actually curious about how the bot would judge my descriptions. Maybe it'd flag them "inadequate" because it notices that the bits of text in the image are not transcribed in the alt-text. Maybe it'd be irritated because I have headlines in my long image descriptions, because they're so long that they need two levels of headlines. Maybe it'd flag them "inadequate" because it goes strictly by WCAG, and a) the alt-texts exceed 200 characters, b) long image descriptions do not belong into the text body by any known official accessibility standards, and c) neither my alt-texts nor my long descriptions are limited to what's supposed to be important within the context of the post.

Anyway, in the meantime, you can follow the account @Alt Text Hall of Fame and the hashtag #AltTextHallOfFame.

CC: @Simon Brooke

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #MastodonHOA #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@-0--1- @David G. Smith Still, first of all, if I posted an image without an alt-text (which I'd never do), AltBot would have to assume full admin rights over the Hubzilla channel that I'm currently commenting from because that's the only way for another Fediverse actor to alter the source code of my posts.

Altering the source code of the post is necessary because Hubzilla, (streams) and Forte neither have a dedicated alt-text field, nor are images file attachments there. Rather, images are embedded directly into the post, in-line, just the same way blogs handle images. And alt-text has to be woven into the image-embedding code in the post. Thus, the post itself has to be altered.

So, assuming AltBot actually manages to circumvent the two most advanced permissions systems in the Fediverse, it would have to trace back an image that it perceives as a file attachment to where exactly the embedding code for that particular image is in the post.

It would have to be able to both understand and write the specific flavour of BBcode used by Hubzilla, (streams) and Forte.

It would have to, for example, take this piece of code...

[zrl=https://hub.netzgemeinde.eu/photos/jupiter_rowland/image/b1e7bf9c-07d8-45b6-90bb-f43e27199295][zmg=800x533]https://hub.netzgemeinde.eu/photo/b1e7bf9c-07d8-45b6-90bb-f43e27199295-2.jpg[/zmg][/zrl]

...and edit it into this.

[zrl=https://hub.netzgemeinde.eu/photos/jupiter_rowland/image/b1e7bf9c-07d8-45b6-90bb-f43e27199295][zmg=https://hub.netzgemeinde.eu/photo/b1e7bf9c-07d8-45b6-90bb-f43e27199295-2.jpg]Digital shaded rendering of the main building of the Universal Campus, a downloadable island location for 3-D virtual worlds based on OpenSimulator. The camera position is about three metres or ten feet above the ground. The camera is tilted slightly upward and rotated slightly to the left from the building's longitudinal axis. The futuristic building is over 200 metres long, stretching far into the distance, and its front is about 50 metres wide. Its structure is mostly textured to resemble brushed stainless steel, and almost everything in-between is grey tinted glass. The main entrance of the building in the middle of the front has two pairs of glass doors. They are surrounded by a massive complex geometrical structure, very roughly reminiscent of a vintage video game spacecraft with the front facing upward. Four huge cylindrical pillars carry the roof end, the outer two of which extend beyond it. All are tilted away from the landing area in front of the building and at the same time outward to the sides. The sides of the building are slightly tilted themselves. In the distance, a large geodesic dome rises from the building. There is a large circular area in front of the main entrance as well as several wide paths. They have light concrete textures, and they are lined with low walls with almost white concrete textures. Furthermore, various shrubs and trees decorate the scenery.[/zmg][/zrl]

Not to mention that AltBot would require extensive detail niche knowledge about the topic covered by the image to be able to whip up the above alt-text in the first place. (By the way: The alt-text example is genuine. I've actually used it. And it's an extremely whittled-down version of the long image description of the same image in the post itself, a description which has to be the longest in the entire Fediverse.)

Ideally, AltBot would do so without flagging the post as edited.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #Hubzilla #Streams #(streams) #Forte #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@-0--1- By the way, I'll accept that AltBot is

AMAZINGLY GOOD

when it's better at describing and explaining images about extremely obscure niche topics accurately than experts on these topics with years of experience.

I've yet to encounter an AI that outdoes my own image-describing in accuracy and level of detail. This, by the way, is likely to require knowledge that only I have.

CC: @David G. Smith

#AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Chris Mills @:neuro: Pixy's Journey :v_bi: Kind of similar here, only that the extra information always goes into the post text itself. That extra information is necessary because I only ever post about extremely obscure topics, and I want people to understand my image posts without having to look anything up themselves.

Whenever I post a wholly original image, I even add two image descriptions, a "short" and purely visual one in the alt-text and an extensive one that includes explanations in the post itself.

And yes, I write my image descriptions myself by hand. I'm on a desktop computer with a hardware keyboard most of the time. Besides, AI can't nearly do what I do.

#Long #LongPost #CWLong #CWLongPost #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@PalmAndNeedle I know. I have evidence.

#AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #AIVsHuman #HumanVsAI

@iFixit

and it doesn't look like you can attach documents to posts

You can't on Mastodon. I could, both here on Hubzilla and on (streams) where I post my images.

But I wouldn't have to. Vanilla Mastodon has a character limit of 500. Hubzilla has a character "limit" that's so staggeringly high that nobody knows how high it is because it doesn't matter. (streams), from the same creator and the same software family as Hubzilla, has a character "limit" of over 24,000,000 which is not an arbitrary design decision but simply the size of the database field.

By the way: Both are in the Fediverse, and both are federated with Mastodon, so Mastodon's "all media must have accurate and sufficiently detailed descriptions" rule applies there as well unless you don't care if thousands upon thousands of Mastodon users block you for not supplying image and media descriptions.

In theory, I could publish a video of ten minutes, and in the same post, I could add a full, timestamped description that takes several hours to read. Verbatim transcript of all spoken words. Detailed description of the visuals where "detailed" means "as detailed as Mastodon loves its alt-texts" as in "800 characters of alt-text or more for a close-up of a single flower in front of a blurry background" detailed. Detailed description of all camera movements and cuts. Description of non-spoken-word noises. All timestamped, probably with over a hundred timestamps for the whole description of ten minutes of video.

Now I'm wondering if that could be helpful or actually required, or if it's overkill and actually a hindrance.

CC: @masukomi @GunChleoc

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #Mastodon #Hubzilla #Streams #(streams) #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #MediaDescription #MediaDescriptions

@Garry Knight @qurly(not curly)joe This, by the way, is something that next to nobody in the Fediverse knows, and that many will deny and fight with all they can:

Alt-text must never include exclusive information that is neither in the post text nor in the image itself. Such information must always go into the post itself. If you don't have room in the post, add it to a reply or multiple.

That's because not everybody can access alt-text. Certain physical disabilities can make accessing alt-text impossible, for example, if someone can't use their hands. Money quote from way down this comment thread:

Deborah schrieb den folgenden Beitrag Mon, 10 Jul 2023 23:30:45 +0200 @jupiter_rowland

I have a disability that prevents me from seeing alt text, because on almost all platforms, seeing the alt requires having a screenreader or working hands. If you post a picture, is there info that you want somebody who CAN see the picture but DOESN’T have working hands to know? Write that in visible text. If you put that in the alt, you are explicitly excluding people like me.

But you don’t have to overthink it. The description of the image itself is a simple concept.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #Fediverse #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #Inclusion #A11y #Accessibility #QuotePost #QuoteTweet #QuoteToot #QuoteBoost

@sunflowerinrain @Tarnport From what I've read, a digital photograph is considered the default. So for brevity reasons, it must not be mentioned.

Any other media must be mentioned, whether it's a painting, a screenshot from a social media app, a scanned analogue photograph, a flowchart, a CAD blueprint, a 3-D rendering or whatever.

But an alt-text must never start with "Image of", "Picture of" or "Photo of". That's considered bad style and a waste of characters and screen-reading time. If the medium is not mentioned, digital photograph falls into its place as a default.

#Long #LongPost #CWLong #CWLongPost #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Alt Text Hall of Fame @David Bloom Yes.

Explanations, or any other information available neither in the image nor in the post text, must never ever go into the alt-text. That's because not everyone can access alt-text. And to those who can't access alt-text, any information exclusively available in alt-text is inaccessible and therefore lost.

#AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Erik van Straten

I often spend more time writing alt text than the text in the toot itself - not just to keep the toot itself as short as possible.

I always spend more time describing my images than writing the post that they go into.

For my meme posts, that's because I have to explain the picture and find the appropriate links to external explanations (KnowYourMeme etc.) to shorten my explanation block if possible.

For my original images, it's because I have to describe them twice. There's always an alt-text which, as of late, fills the 1,500-character limit imposed by Mastodon, Misskey etc. to the brim. But that alt-text is only a shortened and slightly adapted version of an extremely long long description which goes into the post text body and which also includes transcripts of any and all text in the image, readable or not, as well as all explanations which I deem necessary for outsiders to understand the image. Since the images are about an extremely obscure niche topic, this means I have to explain a lot.

A while ago, I spent two full days, morning to evening, researching for and describing and explaining one single image. The result was probably the longest image description ever posted in the Fediverse. And I actually had to limit myself, otherwise the description would have been even vastly longer and taken over a month to complete. Good thing I don't have any character limit to worry about. The only exception is that Mastodon may reject posts from outside with over 100,000 characters.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

@Alan Levine Judging by the advice I've read so far, it's always best to describe the colour using basic colours plus attributes such as brightness, saturation and what other basic colour or colours the colour you describe is leaning towards.

For example, "light, yellowish orange", "a darker, slightly less saturated, slightly more brownish tone of orange", "various shades of slightly yellowish, medium-light-to-medium brown", "a solid, slightly pale medium blue with a minimal hint of green", "a medium-dark wood texture, slightly reddish, slightly greyish". All actually used by me in the long descriptions in (content warning: eye contact) this image post.

If the name of the colour plays a role, use it and then describe the colour in the same way as above. Blind or visually-impaired people may not know what Prussian blue or Burgundy red looks like.

@Stefan Bohacek @Faerie @cobalt @Tanya McGee Wheatley What do you say, is that appropriate, complete overkill or still insufficient?

#Long #LongPost #CWLong #CWLongPost #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta