One interesting development from generative AI is that it has opened a lot more people's eyes to seeing kitsch.
Professional artists generally don't see great value in the style of art exemplified by these AI-powered image generators. It tends to be formulaic, derivative, and melodramatic with all the high-contrast portraits and sunsets and oversaturated pastoral landscapes. But popular taste has been in many ways stuck a century or more in the past.
Now that these styles can be generated at the click of a button, it's far easier for even laypersons to see them as essentially canned forms of expression. I wonder if this augurs a larger shift in artistic taste. The introduction of photography fundamentally changed what people want from human-made images, and AI could have almost as far-reaching effects.
I think I generally agree with this take, with the caveat that I still hugely value AI generated art, even the kitsch stuff. Why? Because it's me. I wouldn't have been able to have a portrait of my wife and I together without finding an artist and commissioning a painting. Now, I can have something that is personalized, pleasant to look at, and meaningful. Even if it isn't an original style or anything noteworthy artistically, I hugely value the AI art because it can give me something that I want, that is meaningful to me.
You're conflating "art commonly created by these AI-powered image generators" with "art that AI is capable of".
Just because most people create things that you categorise as kitsch should not fool you into thinking that's an inherent limitation. There's almost no limit to the kind of images that can be created given the right input.
Unless you want to define kitsch as "anything that an AI image generator is capable of producing"?
It’s the opposite really. I’m not claiming anything about AI image generators. I’m saying that people have quite limited tastes, and so they end up going for kitsch because it looks good at first glance.
But now that it’s so easy to generate, will it mean a shift in taste? The current crop of AI art looks the same because enough people think those limited styles are good. But in ten years, users may be a lot more demanding.
Think about 1950s pop music, before the invention of electric guitars and synthesizers and sequencers. The range of musical expression is quite limited to modern ears. But at the time, it was uncontested among ordinary listeners that “good music” had that particular sound. There was an enormous gulf between avant garde contemporary music and pop music, just like in art today.
You are absolutely right regarding the capabilities.
Although many image generators now seem to be fine-tuned to output impressive but formulaic art by default, to ensure a good first-impression for a broad audience.
Midjourney for example has always been great at generating very "cool" art easily, but seems to be less versatile and always tends towards that same people-pleasing style.
The style associated with AI image generators is by no means all they are capable of, it’s what is both actuallypopular in use and promoted by vendors because they expect it to be popular and blandly inoffensive.
Also there's nothing inherent in currently popular style that makes it “canned expression”; is used for canned expression because it’s popular; when popular tastes shift, so does canned expression; that's how commercial vs. fine art moves.
If AI changes the details of that other than accelerating the process, it’ll probably be by a “Latinum is valuable because you can't replicate it” effect, which probably won’t hit exactly where what current professional artist see as most interesting is any more than it will hit where current populat art is.
> I wonder if this augurs a larger shift in artistic taste.
People have distorted/blurred/abstracted images from the start with 'Vaseline on the lens' to airbrushing to modern concerns about 4k showing too much detail. Popular art wants to remove the detail and portray the subject as better-than.
I'll note that it's AI art generated by non-artists that tends to be this way (the majority of AI art) but it's definitely possible to get much more human looking output with proper conditioning/model tuning. I do think you have a point though.
I think you're absolutely right that generative AI art doesn't produce any "great value" at all, in terms of Art with a capital A.
But on the other hand, probably 99% of what professional illustrators do isn't, and doesn't need to be, great Art either.
They're just trying to quickly produce a hero illustration for an article, or turn an author headshot into a simpler sketch, or add some appropriate clothing to a background NPC. Being "formulaic" and "derivative" and a "canned form of expression" here is a feature, not a bug.
Which is why generative AI seems so great to me.
In terms of "kitsch" -- yeah I'd never in a million years hang something AI-generated on my wall as if it were fine art produced by a human soul trying to express something.
And I think the creator of Drawbert should focus entirely on this being a tool for being professional portraits for business and branding use, rather than "stunning portraits" or "masterpiece paintings". The former is what it actually is, the latter it ain't.
I don't know if I'd call the portraits "stunning" but this is exactly the type of image that is extremely useful to use as an author portrait on a blog or publication, in a list of employees on a startup's team page, and so forth.
You can compare with Wall Street Journal hedcuts [1] or the portraits in the Economist's "By Invitation" [2].
Because actual photos often reveal "too much", with a lot of distracting detail not relevant to your "essence", and you're never going to get a series of photos of different people to look all that consistent unless you're using the same photographer with the same setup in the same location, which usually isn't practical.
So this kind of tool that accurately reproduces someone's face and expression and hair accurately, but stylizes the rest, is actually super useful in communicating just the right amount of visual information in a lot of professional contexts.
So kudos to the creator -- I think you might be able to turn this into a business if you market this to businesses that want to use images of their employees, where the business pays to decide on a particular consistent "look", and then can just feed in and update images of employees/reporters as needed. Charge something like $199 to define the look, and then $99/yr for portraits based on up to 100 original photo uploads per year, $179/yr for 200 photos, etc.
Cross-Origin Request Blocked: The Same Origin Policy disallows reading the remote resource at https://drawbert.ai:1000/convert. (Reason: CORS header ‘Access-Control-Allow-Origin’ missing). Status code: 502.
I’ve deployed something similar for Instapainting using convolutional texture nets: https://www.instapainting.com/assets. This is some 5+ year old stuff but with hand-picked models still stands up against newer stuff albeit limited to select styles.
I assume this is using newer generative techniques like stable diffusion.
Well it has been almost 24h now, and it is still not up.
I think it is concerning that you did not put up a simple "we immediately delete the images you upload and we do not store any data about you, nor do we track you using cookies or share data with third parties".
Since you do not have a privacy policy, I would expect that you do not do any of these things. So why not write this?
I tried it with some photos of my partner and I - whatever the photos I choose or how many times I tried, I'm always very realistic but my partner is ... somebody else. I guess your training set lacks some particular faces or something, because it always want to render someone else, and curiously it's very coherent in that someone else.
It's nicely presented, good job! Unfortunately with my free credit it came up half baked, as grayscale with some color splotches and without the background I described. The caricature rendition of my supplied photo is very good.
Will we have the tendency to steer away from generated content because it will become « cheap » ?
Or will those generated painting will retain the same value as current paintings done by a human in their appreciation and value ?
if you want to do this yourself you're going to want stable diffusion and one of the clipspace face embedding controlnets (tencent Photomaker is one), the ones that keep the input pose exactly are likely using depth as well or maybe straight image2image, I demoed something like this at open sauce this weekend making pixel art on an ancient dot matrix printer.
I'll endeavor to package everything up and post a repo in the near future, this sort of low effort sass stuff really irritates me. I'll do my best to replicate this functionality in a easy to use open source UI/UX
> this sort of low effort sass stuff really irritates me
Why? There's a big world out there with people who don't know any of these words (controlnets, stable diffusion, open source,...) and it's probably targeted at these.
Not even close. Traditional Photoshop filters can't do anything like this.
I know the filters you're talking about, and they can certainly turn photos into some faux-artistic effects, but the result is really bad, and everything in the image remains in the exact same place.
This is applying an actual artistic style, and also changing body position, background, hair, lighting, etc.
So I don't know why you'd want to be so dismissive of something like this.
To an untrained eye, it all looks the same really. There is no artistic style here that you can clearly see in any decent museum of fine arts. It really does just look like photoshop filters.
Not sure the power costs of running a model like this over the caloric costs of a human + the power to power their photoshop machine, for instance, is worth it.
Lots of layers and lots of mattes. You take any picture of a person, and matte them out of the background. You then find a background you like, and arrange as you see fit. You then find whatever hair style you like, and matte it in as well. Plenty of other filters involved as well. Hell, you can even play with this stuff in Illustrator to get the "painted" look by converting to vectors.
How many magazine covers do you see and actually believe that was anything close to what the original image looked like? I have no idea how long this thing takes, but I'm guessing it's helluva lot faster than a person in PS, and it will be doable with anybody with zero amount of talent required.
To think it couldn't be done in PS is just lack of knowing good PS artists.
One interesting development from generative AI is that it has opened a lot more people's eyes to seeing kitsch.
Professional artists generally don't see great value in the style of art exemplified by these AI-powered image generators. It tends to be formulaic, derivative, and melodramatic with all the high-contrast portraits and sunsets and oversaturated pastoral landscapes. But popular taste has been in many ways stuck a century or more in the past.
Now that these styles can be generated at the click of a button, it's far easier for even laypersons to see them as essentially canned forms of expression. I wonder if this augurs a larger shift in artistic taste. The introduction of photography fundamentally changed what people want from human-made images, and AI could have almost as far-reaching effects.
I think I generally agree with this take, with the caveat that I still hugely value AI generated art, even the kitsch stuff. Why? Because it's me. I wouldn't have been able to have a portrait of my wife and I together without finding an artist and commissioning a painting. Now, I can have something that is personalized, pleasant to look at, and meaningful. Even if it isn't an original style or anything noteworthy artistically, I hugely value the AI art because it can give me something that I want, that is meaningful to me.
> pleasant to look at
That is very subjective. Mass market art loses appeal fast, popular turns common, what is unique steals the spotlight and tastes change.
I can see best quality AI generated custom portraits being as pleasant to look at as selfies.
You're conflating "art commonly created by these AI-powered image generators" with "art that AI is capable of".
Just because most people create things that you categorise as kitsch should not fool you into thinking that's an inherent limitation. There's almost no limit to the kind of images that can be created given the right input.
Unless you want to define kitsch as "anything that an AI image generator is capable of producing"?
It’s the opposite really. I’m not claiming anything about AI image generators. I’m saying that people have quite limited tastes, and so they end up going for kitsch because it looks good at first glance.
But now that it’s so easy to generate, will it mean a shift in taste? The current crop of AI art looks the same because enough people think those limited styles are good. But in ten years, users may be a lot more demanding.
Think about 1950s pop music, before the invention of electric guitars and synthesizers and sequencers. The range of musical expression is quite limited to modern ears. But at the time, it was uncontested among ordinary listeners that “good music” had that particular sound. There was an enormous gulf between avant garde contemporary music and pop music, just like in art today.
You are absolutely right regarding the capabilities.
Although many image generators now seem to be fine-tuned to output impressive but formulaic art by default, to ensure a good first-impression for a broad audience.
Midjourney for example has always been great at generating very "cool" art easily, but seems to be less versatile and always tends towards that same people-pleasing style.
1 reply →
The style associated with AI image generators is by no means all they are capable of, it’s what is both actuallypopular in use and promoted by vendors because they expect it to be popular and blandly inoffensive.
Also there's nothing inherent in currently popular style that makes it “canned expression”; is used for canned expression because it’s popular; when popular tastes shift, so does canned expression; that's how commercial vs. fine art moves.
If AI changes the details of that other than accelerating the process, it’ll probably be by a “Latinum is valuable because you can't replicate it” effect, which probably won’t hit exactly where what current professional artist see as most interesting is any more than it will hit where current populat art is.
> I wonder if this augurs a larger shift in artistic taste.
People have distorted/blurred/abstracted images from the start with 'Vaseline on the lens' to airbrushing to modern concerns about 4k showing too much detail. Popular art wants to remove the detail and portray the subject as better-than.
I'll note that it's AI art generated by non-artists that tends to be this way (the majority of AI art) but it's definitely possible to get much more human looking output with proper conditioning/model tuning. I do think you have a point though.
I think you're absolutely right that generative AI art doesn't produce any "great value" at all, in terms of Art with a capital A.
But on the other hand, probably 99% of what professional illustrators do isn't, and doesn't need to be, great Art either.
They're just trying to quickly produce a hero illustration for an article, or turn an author headshot into a simpler sketch, or add some appropriate clothing to a background NPC. Being "formulaic" and "derivative" and a "canned form of expression" here is a feature, not a bug.
Which is why generative AI seems so great to me.
In terms of "kitsch" -- yeah I'd never in a million years hang something AI-generated on my wall as if it were fine art produced by a human soul trying to express something.
And I think the creator of Drawbert should focus entirely on this being a tool for being professional portraits for business and branding use, rather than "stunning portraits" or "masterpiece paintings". The former is what it actually is, the latter it ain't.
These are actually really, really useful.
I don't know if I'd call the portraits "stunning" but this is exactly the type of image that is extremely useful to use as an author portrait on a blog or publication, in a list of employees on a startup's team page, and so forth.
You can compare with Wall Street Journal hedcuts [1] or the portraits in the Economist's "By Invitation" [2].
Because actual photos often reveal "too much", with a lot of distracting detail not relevant to your "essence", and you're never going to get a series of photos of different people to look all that consistent unless you're using the same photographer with the same setup in the same location, which usually isn't practical.
So this kind of tool that accurately reproduces someone's face and expression and hair accurately, but stylizes the rest, is actually super useful in communicating just the right amount of visual information in a lot of professional contexts.
So kudos to the creator -- I think you might be able to turn this into a business if you market this to businesses that want to use images of their employees, where the business pays to decide on a particular consistent "look", and then can just feed in and update images of employees/reporters as needed. Charge something like $199 to define the look, and then $99/yr for portraits based on up to 100 original photo uploads per year, $179/yr for 200 photos, etc.
[1] https://www.wsj.com/articles/whats-in-a-hedcut-depends-how-i...
[2] https://www.economist.com/by-invitation
Thanks for adding this angle, definitely not something I thought about, could be a good use case for the future path of this tool
The POST to https://drawbert.ai:1000/convert returns
Cross-Origin Request Blocked: The Same Origin Policy disallows reading the remote resource at https://drawbert.ai:1000/convert. (Reason: CORS header ‘Access-Control-Allow-Origin’ missing). Status code: 502.
after ~2 mins
I tried reading the terms of use and privacy policy. Both lead to 404 errors (https://drawbert.ai/terms and https://drawbert.ai/privacy)
I’ve deployed something similar for Instapainting using convolutional texture nets: https://www.instapainting.com/assets. This is some 5+ year old stuff but with hand-picked models still stands up against newer stuff albeit limited to select styles.
I assume this is using newer generative techniques like stable diffusion.
Interesting to know how things were done before the generative AI wave, and yes this is based on stable diffusion
Privacy policy returns 404
It is important in service like this.
Sorry I was rushing to get the site live, privacy page will be up soon
> privacy page will be up soon
Well it has been almost 24h now, and it is still not up. I think it is concerning that you did not put up a simple "we immediately delete the images you upload and we do not store any data about you, nor do we track you using cookies or share data with third parties".
Since you do not have a privacy policy, I would expect that you do not do any of these things. So why not write this?
Why the rush?
7 replies →
I tried it with some photos of my partner and I - whatever the photos I choose or how many times I tried, I'm always very realistic but my partner is ... somebody else. I guess your training set lacks some particular faces or something, because it always want to render someone else, and curiously it's very coherent in that someone else.
I'd suggest using a different photo of your partner, something that represents them better. That makes a difference in my experience
I tried 4 different photos. Oh well, thanks anyway.
It's nicely presented, good job! Unfortunately with my free credit it came up half baked, as grayscale with some color splotches and without the background I described. The caricature rendition of my supplied photo is very good.
Thanks for the feedback!
Will we have the tendency to steer away from generated content because it will become « cheap » ? Or will those generated painting will retain the same value as current paintings done by a human in their appreciation and value ?
eventually, the novelty will wear off.
if you want to do this yourself you're going to want stable diffusion and one of the clipspace face embedding controlnets (tencent Photomaker is one), the ones that keep the input pose exactly are likely using depth as well or maybe straight image2image, I demoed something like this at open sauce this weekend making pixel art on an ancient dot matrix printer.
I'll endeavor to package everything up and post a repo in the near future, this sort of low effort sass stuff really irritates me. I'll do my best to replicate this functionality in a easy to use open source UI/UX
> this sort of low effort sass stuff really irritates me
Why? There's a big world out there with people who don't know any of these words (controlnets, stable diffusion, open source,...) and it's probably targeted at these.
I would be interested in the repo.
Could really use a $3 or $5 credit pack. $10 is a bit big to jump into without trying it out.
there are some free credits to try things out if you sign in
Sign up link sent per email redirects to localhost
thanks for reporting this, it's now fixed
Seems nice. I'd use if it was a GIMP plugin.
These look great! Can you explain a bit about the underlying technique? Is this using InstantID?
Okay this is sweet. My wife is gonna love this lol.
[flagged]
[flagged]
Examples look like pretty basic Photoshop filters circa 20 years ago.
Not even close. Traditional Photoshop filters can't do anything like this.
I know the filters you're talking about, and they can certainly turn photos into some faux-artistic effects, but the result is really bad, and everything in the image remains in the exact same place.
This is applying an actual artistic style, and also changing body position, background, hair, lighting, etc.
So I don't know why you'd want to be so dismissive of something like this.
To an untrained eye, it all looks the same really. There is no artistic style here that you can clearly see in any decent museum of fine arts. It really does just look like photoshop filters.
Not sure the power costs of running a model like this over the caloric costs of a human + the power to power their photoshop machine, for instance, is worth it.
2 replies →
Genuine question then, how do you do something like this with PS or Gimp?
Lots of layers and lots of mattes. You take any picture of a person, and matte them out of the background. You then find a background you like, and arrange as you see fit. You then find whatever hair style you like, and matte it in as well. Plenty of other filters involved as well. Hell, you can even play with this stuff in Illustrator to get the "painted" look by converting to vectors.
How many magazine covers do you see and actually believe that was anything close to what the original image looked like? I have no idea how long this thing takes, but I'm guessing it's helluva lot faster than a person in PS, and it will be doable with anybody with zero amount of talent required.
To think it couldn't be done in PS is just lack of knowing good PS artists.
2 replies →
photoshop requires human work, but this is fully done automatically, that might be the major difference
Just requires a single snipped of javascript to do the same automatically. Photoshop supports robust scripting.
You can also do much the same with imagemagic, no AI required.
3 replies →