I think this thing is just trained to like large hero images. Tried some of the most beautiful sites I know and got scores in the 10s and 20s, presumably from the presence of text.
Then I tried a number of sites with worthless huge hero images and got scores in the 80s.
I mean like all ai, take this with a grain of salt, but I think there is an obvious bias in this to the 'giant hero image that goes all the way to the fold' design that a lot of low effort start-up websites use.
Agreed. I tested two of my personal websites[1][2]. The one with the text-heavy poetry scored 38% with a "meh" comment. The one with the gratuitously oversized banner image got 72% with a "oh yeah!" comment.
It does have an AI running the analysis. I like my poetry website. Maybe the AI was trained on good poetry and reacted poorly to my stuff?
If it likes hero images, the black background might look suspiciously like an image to the algorithm if they haven't explicitly done something to ensure similar results on color inversion.
This sounds about right. I got dinged a lot on my website for "Visual Clarity" although imho it's a pretty standard blog page with sidenotes, and the "similar scores" all had a decent amount of text.
Looking at the "similar scores" carousel for higher scores was primarily huge hero images with little text.
I don't find your site's links clear at all by the way. It turns out that emojis are supposed to indicate actions/links, but until that's common you might want to give link text a slightly different color or maybe underline it.
Not sure I hover over what seems to be regular text on a typical website. On desktop you need to put your cursor on it and on mobile your finger, not sure it's that different.
My website's homepage which consists of a 128x128 picture of me, text and a table got 54. I guess the mere presence of a image made the AI like it a bit more.
Hi, Thanks for your feedback you are right. Currently algo evaluate how a user will feel...when they look at your site above the fold...it's like tinder you just look at the pic and decide to swap right or left.. this is based on : Users make lasting judgments about a website’s appeal within a split second. This first impression is influential enough to later affect their opinions of a site’s usability and trustworthiness...of course lot of areas we can improve...hope u enjoyed..
I think your project will not be taken seriously by the target audience if you imply that the AI isnt extremely biased, as any "pretty or not" judgement.
Providing a "what people might like better statistically" service is interesting, claiming you have a "the AI knows how the user will feel" algorithm is not a good strategy
By improving over time do you mean training the model for more epochs or having some kind of feedback loop that tells the algorithm whether the score was accurate or not based on human input?
I think the later makes more sense. Some kind of input from the users like "I actually liked the website visually" when the algorithm gives a low score and vice-versa would help the model be more accurate?
The results under "Here are few websites with similar visual scores as yours" were especially amusing for my site.
I've received a score of 26 and the similar websites were:
- Generic unstyled XML Access Denied document.
- Google account log in page.
- Generic unstyled "Your request has been blocked" text.
- "Please verify you're a human" HTML error message.
- Random JSON error message.
I entered my personal website, https://merely.xyz . That website is served with a single request; perhaps it's been classified as an error page because of that?
I tested https://gildas-lormeau.github.io/ (score 56) which is also a single request website. Almost all the websites with similar visual scores are okay for me.
I also got a 26 for a site I entered. It was a bit confusing because 26 happens to be a fixed point, the site gives you a score (26% of users will like your site) and also ranks you among all the sites people have entered (you scored higher than 26% of all sites entered).
Agreed. I put together a little library to handle denying access to "local" resources, to abstract this a little and avoid overlooking common mistakes:
Good ol' metadata strikes again. I wish AWS would make a virtual block device for accessing this content instead of network (at least as an option). Much easier to protect at the OS level.
The v2 service helps with this but it doesn't beat awareness of the issue.
Complete nonsense in the absence of any sense of why a site is designed the way it is. Is it a product brochure? Is it something functional (e.g. a search engine)? Does it target the general public? Or (say) highbrow art appreciators? Or tech geeks? In each case the aesthetic might be deliberately designed to attract/repel according to target audience. There is/can be no generic "attractive".
I'm not very convinced of this AI. I got 96% for http://peppermind.com and only 68% for https://talumriel.de - I'm almost certain that most users would prefer talumriel, and both scores them seem overrated to me. I mean, I really have no clue about design...
Thanks for this really. I looked it and trust me AI has got this completely wrong ..we will include this as exception when we train again.. thanks for trying..
My default Apache 404 page got a 38% of users will like it, 26 points for visually average, 20 for visual and 30 for clarity. After a few different URLs, my fail2ban kicked in, and the site complained about the URL being invalid. Fair enough.
Hilariously my blog (kn100.me) also got 38 percent and the site provides screenshots of other sites with similar scores, which were all 404/500 error pages... I thought my blog looked nice :(
Using my somewhat powerful but also flawed neuro net trained on over 20 years of intensive web usage gives your blog design a thumbs up. One suggestion is adding padding-top: 1em to the "Posts" heading to make it more clearly separated from the quick summary above it and more clearly related to the content below it.
Most of the page contains punctuation errors, even in the title.
Remove the space before the question mark.
In English, it is always an error. There should be no space between a sentence and its ending punctuation, whether that's a period, a question mark, or an exclamation mark. There should also be no space before a colon, semicolon, or comma.
I dont want to sound like a hater here, but why is it that so many people in this thread claim a 72% score?
This whole thing seems to me like it isn't any AI by the common understanding, but just a few hardcoded scores and criteria.
Now, thats probably wrong, and the AI is super neat, but I think it doesn't look very good when it does that.
I dont think the chances are high that so many different websites all score exactly 72%. Not 71, not 73, 72.
Its weird. Now the whole site looks nice and all, I saw the papers referenced, too.
But again, when I read that it's "97% accurate", I wonder how that is even measured? I understand that there are measures of how easy to use a site is, etc. So I wonder what cases those 3% were, where your AI was "wrong"? Did you hand-review them, or what?
Thanks for trying out. yes you are right. Tips are not upto the mark. Our algo tries to put a site in 8 buckets -b,..+A and provides tips. So it's not that accurate when evaluating 50% versus 53% but mostly accurate with difference of over 10%.. so 50% won't be same as 60% for example.. of course it is based on limited amount of data we had to guess it..point taken.. we really appreciate your feedback and will incorporate.. cool.
What's it actually doing under the hood? Just taking a screen grab of the first few hundred vertical pixels x the full width? What's the actual resolution that it's grabbing, and how does the algorithm work and come up with scores? Is the system backed by some sort of data from a representative group of users?
More like generate a random number from 30 to 90 and show gibberish data, cool idea though to fool small businesses into improving their visual score and charging money, good gimmick though
Individual metrics 60/60 means u have fine balance...Algo keeps on refining as it processes more data..scores are just indicative not to taken as absolute basis. Thanks for trying.
Thanks! https://doomlaser.itch.io has a bunch of the free ones. It gets "better than 84% of millions+ websites analysed by Visual MIND AI", with a 90 in "Visual Appearance", whatever that might mean :)
I think I stumbled upon your itch.io page at some point because I remember Shit Game, but never got a chance to really play it, hopefully I can get to play some of these this time, nothing can go wrong with that 90 visual appearance score!
Ah, cool. I think it still holds up. I put some heart into it, anyway.
Of my free games, the most personal and artsy is probably Standard Bits. It premiered at Gamma 256 in Montreal next to Jason Rohrer's Passage, and people have said some nice things about it over the years: https://doomlaser.itch.io/standardbits It takes place on a 100x54 pixel field, blown up to full screen, and you control a single pixel on a journey. Named after the low-level pixel blitter in the original Macintosh toolbox API..
For some reason this one won't start up on my macOS, gave some error dialogs on first time but I didn't get a chance to see what's exactly said, looks like a wine issue though. However I really like what I saw from the short video, I can smell some alternative / interesting gameplay there, it reminds me of some of Michael Brough's games. (I'll try download the windows version and use the maybe newer version of Wind on my system)
> Visual Mind is an AI engine specifically designed for understanding and scoring visual appearance of a website. Visual Mind has analyzed over a million websites to achieve an accuracy rate of over 97%.
How does accuracy work in a project like this, where the result is subjective?
“You can improve visual score by using better images and improving site layout to make it little more denser.”
These suggestions need to be improved if they are going to be the output of this tool.
Better images? Define better. It doesn’t ask what the images or site are trying to achieve. Maybe for their intended purpose these images are the best.
And making it more dense? Again, why?
If you’re going to give design feedback it might be valuable to consider what valuable design feedback looks like. There are books on this.
Agree that recco are not that upto the mark at the moment as it is just the first version...hence we called it tips. I appreaciate your feedback and thanks for trying ..
Amusingly, from the comments here, it seems like everyone wants some kind of "HotOrNot" service for websites. Crowdsource the ratings! Then you can drop the "A" from "AI" :)
Yeas your's is militaristic...may be the algo though you have too much text..and no images so low visual appeal..may be a case we need to consider in the next training set..don't take the recommendation from algo on face value ..but just as a overall guidance..thanks for trying and sending over feedback.
We use Bootstrap at CareDash with larger hero images on landers with clear calls to action. This structure tends to perform well on A/B Tests too as it leads the user to the action meant to be taken.
I gave my website without the login credentials and got 38% for the 401 page. Then I added the credentials and got the same exact percentage. I know the creds worked, because the screenshot changed. I'm not going to say this assessment site is completely useless, because the colorized screenshots at the bottom are pretty cool and give me more theme ideas, but other than that...
And I can't even save the images, because they are just all the same screenshot, base64-embedded in the HTML and then altered with JavaScript-based filters! What is that!?
Score is really base don the screenshot generated. If something is hidden behind the login system can't read it so will get error. We may in future have a functionality to upload ur own screenshot instead of taking it by machine. Thanks for the feedback.
> Visual Mind summary report for http://usecastup.com - 72% of users are expected to like your site. Your site is visually awesome. You can further improve it by using better images and considering little denser layout.
Well, although I'm happy with the comment, that "72%" seems made up entirely.
@myraahio Given how quickly your site runs, I'm not entirely sure it runs a neural network - seems more like some weighted analysis based on some defined measures on simple algorithms? I'll take a wild stab and guess that you're doing some analysis, some manual labeling and k-means clustering?
I would really consider running a light weight network over a small image (down-scaled to something like 320x200), if you're only binning classes then you should be able to get it to <1ms on a low-power CPU. Given how much web pages vary, you'll likely need upwards of 10k training images.
Consider having some feature on your site to allow users to help rate pages (with moderation oversight).
It's a good start anyway! I look forward to seeing this in the future!
Edit: Also consider some expansion opportunities - rating screenshots of software interfaces, mobile phone apps, etc, etc.
@bArray really appreciate your feedback. Especially expansion opportunity.
DL model is trained on the GPU - heavy AWS instance..end point rests on smaller instance. Sometimes it gives faster results when you run a URL which it has already processed. It will happen in case of famous sites. For new sites it need to process the info so it takes about 10 seconds to create report.
I love the idea but it is of limited use to me right now:
How do I know this metric is any good?
What does “attractiveness” mean and what desired outcomes does it correlate to?
For example HN is ugly (sorry for the candor!) but it’s a plus - it’s a statement about what you write being most important and also by not changing the ui it’s lovely to use. Now compare that to Reddit which might come though as a prettier site at first glance but bitching about the Reddit ui and then someone mentioning old.reddit and friends is a trope: lots of people hate the Reddit ui. So on some metric that matters to me HN wins.
Finally - if I score X how do I actually improve the score?
I usually visit page performance sites to get some tips on things to improve on the performance, I think that this is a great idea to have a score on the appearance as well.
I got "You can further improve it by using better images and considering little denser layout." as a tip to improve, it would be great if there could be more details on the feedback.
Very cool idea but it just does not seem to work just yet. My blog got 26 which I think looks quite nice (http://granitosaurus.rocks) but some random color rainbow websites score double that (like already pointed out: http://dokimos.org/ajff)
Are there any visual design scoring systems that somewhat work?
https://datacrayon.com gets a visual appearance score of 90, but a visual clarity score of 20. "60% of users are expected to like your site"[1].... interesting
Mine is supposed to be good looking (score 90), but too complex (score 20). It’s a simple content in the middle + sidebar layouts. Indeed, very complex.
Beauty is in the eye of the beholder as they say. One big missing piece here is context. Attractive to whom? And in what context?
My site is all about minimalism - https://www.anders.co/ - it got an average score. And I guess that is fair. It was also compared to google.com, which I would say is anything but average.
I achieved 60 (72% of users will like it) with my side-project[1] which is mostly an artwork, I believe its colorfulness helped the score a lot. I'd love for the algorithm to give me clear indications on what it believes I should improve, e.g. make this button wider, make the image bigger.
its definitely just a handful of scores and a handful of comments that get picked rather randomly. Just count the amount of "72%" posts in the comments here
I don't know... my website looks "average" and only 30% will consider my website clean? My website is so minimalistic, that I think the score should be higher.
I'm using the publii default theme on my website and it ranks quite high. It said to get an even higher ranking I should consider denser content, which makes sense given I just moved to publii and it's a bit light right now.
I'm getting visually stunning for https://msb.com but I'm confused does it look only above the fold or the whole site? The screenshots only seem to show the top of the website
Something must have went wrong. can you provide the site link. sometimes if some space is after the url while copy pasting can cause this error. Thanks for trying.
This algo only rates sites on how good visuals are not on utility. Google is a site where you go primarily for utility but ALGo says if you make your site like google only 38% of people are expected to like it.
yes u are right and thanks for the feedback. It's hard to say if you score is 72 out of 100 is it good very good or excellent.. we are just following what happens in the exam so 70% is good enough..not to be taken literally...moral of the story if score is above 60% the site is good...not to be taken as negative...
I guess this sort of depends on the sample set. If it was fed all webby award winners and sites from dribbble then scoring a 51 would be somewhat amazing. An exam has an absolute range given an absolute understood input. I suppose the weird thing here is that we don’t know the inputs. If mainly HN and UX conscious forms have been using the system then 51 is also good.
I don’t have an answer here but it got me thinking. Thanks for this demo! My company site scored a 72 and I might be happy about that.
All I'm getting a generic info page without scores, and when I click the button to show my score, it shows "Error posting feed", taking me to the initial page. Is the app broken, or am I missing a catch?
It seems like for this to be successful you'd want access to engagement/conversion data to train your model on. Google / Facebook would be positioned to execute on that, others less so.
Why do people forget that AI is just a buzzword for some kind of pattern matching ? So it's obviously NOT able to evaluate "creative" capabilites because... well... creativity is about being DIFFERENT and not the same.
Moreover, a site like Facebook is attractive NOT because of its design (mainly) but because of the service it provides. So a Facebook designed clone will have no interest, even if it looks like Facebook.
To me, it looks like it's just the wrong tool for the problem to be solved
> So it's obviously NOT able to evaluate "creative" capabilites because... well... creativity is about being DIFFERENT and not the same.
Neural nets can approximate any function as accurately as you'd like. Blindly shoving whole sites prelabled by their creativity into a convnet seems likely to end poorly, but that doesn't mean a better training method can't be applied which more faithfully represents the problem domain (no clue what the inside of this particular tool looks like, just commenting on the general infeasibility claim).
> So a Facebook designed clone will have no interest, even if it looks like Facebook.
Doesn't this site just claim that appearances also matter, not that a polished turd will be successful?
> just commenting on the general infeasibility claim
I agree: the main problem of "AI" is the knowledge base used. And the biais associated. Which mean that you can correct any biais by changing the reference... but introducing other biaises.
It would be possible to use such a tool either to focus on smaller tasks (let's say: focusing only on photo gallery of professional photographer websites) and be more specific (but less creative) or to grow the knowledge base to include paintings, architecture, object design to exact more general design principles (gestalt and so...)
In the end, the "AI" will be to create new "mixes" of different already used concepts... but I don't see how it would be able to create new concepts. The "AI" will - as much as I understand the technology - stay inside the space defined by it's knowledge base. If all the website of the knowledge base have only white or black background, the "AI" can't "think" to use a green background, because it doesn't have any inference mecanism to think of the background color as any color. It is limited to the background that was fed.
> In the end, the "AI" will be to create new "mixes" of different already used concepts... but I don't see how it would be able to create new concepts. The "AI" will - as much as I understand the technology - stay inside the space defined by it's knowledge base. If all the website of the knowledge base have only white or black background, the "AI" can't "think" to use a green background, because it doesn't have any inference mecanism to think of the background color as any color. It is limited to the background that was fed.
"AI" that does what we want is limited to whatever rules we impose on it. For a lot of problems the most efficient way to impose rules is to provide a set of samples and interpolate, but if we have some way to meaningfully define creativity (which I don't think will be feasible in general any time in the near future, maybe ever) then we can produce an architecture which matches that definition (and if we're hung up on the generative portion of that, a trivial though expensive way to accomplish generation is to enumerate outputs and check if they match our definition for creativity).
"AI" isn't limited to the samples it's given; it's limited to the biases we impose. We can explicitly impose a bias that says hue matters if we so desire.
IMHO, "creativity" is all about breaking existing rules, replacing them partially or totally with others. In math, it's a new axiom set, allowing/forbidding new inferences. And it's also using analogies with other domains to find new intuitions and new deductions.
However, the main problem with creativity is not reducing the knowledge base, it's finding new knowledges to extend the knowledge space consistently. Sometimes enumeration or automatic generation can help... but in that case, it's "only" a fixed set of meta-rules
When u meet someone all the time u start liking it. Our tool helps u get an average output that if u see someone for the very first time how much attractive u will find them. of course AI is not absolute but gives u some data points to u to decide. Thanks for your feedback.
Actually, it will give some feedback... but will it give any VALUABLE feedback ? I mean: how is it better than any website scoring by Google, Yahoo and so ? How is "AI" better than some UI/UX human research and thinking ?
It looks like everybody is more or more tired of all the websites looking the same because of the "bootstrap"/"material" design. Your tool will only score any website against the tiring majority, so in fact I'm really not sure that it's such a good idea.
I think that a better - but really far more difficult - task would be to different and usable. But you can't achieve this only with "AI" on website: you need to add also some kind of graphic art knowledge and UX principles. Example: instead of considering a color scheme against what is used all over the place (hint: default bootstrap or default material), you should check color scheme against paintings or photographies, and score better what is "beautiful" AND original. Maybe you could add for example some link to kulers and that kind or color scheme design tool. THAT would make it a VALUABLE tool, because it would help the website to stand out and be usable...
Input: https://www.wikipedia.org/ Output: 38% of users are expected to like your site