What’s interesting is even in human-generated art, hands are usually considered the most difficult to draw/paint, so I am comforted knowing AI can’t master hands either LOL
i love how Bernadette's aesthetic has gradually shifted from poised but meme-y victorian millenial to just. accepting that she has the power to do whatever the fuck she wants and nobody will stop her
Mildly disappointed that The Dating Game didn't end with you having to choose a very cursed individual to have a candlelight dinner with, but overall this video was hysterical and made my day.
I just realised how these AI generated images could potentially confuse things in the future when it comes to historical research, if any of these images get saved and misplaced and miscategorised by people not knowing better or just by mistake. It likely wont happen any time soon, except perhaps in armature circles, but as time goes on if care isn't taken then I wonder if the odds of this getting mixed up could grow.
There is a lot of bad stuff that can happen and the better they get the worse it will be. Anyone could make a picture of you kicking a dog, or of you naked (that is already happening), or any sort of fake news photographs that could sow misinformation.
I don't think so. I mean, people continue to paint knock-off "historical" portraits (of celebrities, for instance) and I'm not aware of any of those ever getting confused for an actual historical piece. And researchers in the future, with AI being more widespread by an order of magnitude presumably, will be hyper-aware of that potential pitfall. I think the larger issue is going to be with intentional falsification of history rather than with the accidental kind.
In school you’re taught proper research practices, these kind of things would pop up on google image search or Wikipedia, not on the Smithsonian website or other credible sources
The thing with AI is it's not actually "artificial intelligence", it's just imitative programs. It's of course going to pull some elements of historical accuracy like silhouette, textile details, etc because it is directly copying from historical portraits from the eras you are specifying. The problem is it's not actually "intelligence", and it has no ability to actually understand what it is copying and why, so that's why you get details like a waist girdle being turned into a necklace.
if it were directly copying it'd be way better at it than what this is showing. What its actually doing is trying to come up with its own map of context based entirely on the massive amount of images its been given. it takes a random mix of pixels and then runs it back through its own image identifying algorithm searching for which change in the pixels brings it closer to pinging as the prompt query. So its moving pixels around until it gets a vague "historical portrait of a woman in the 1570s" vibe off what its made. if it were just cutting bits of actual portraits and shuffling them around, you'd expect ruffs that look like ruffs not vaguely approximately ruff suggesting pixels
Well, this type of AI program is "intelligence." But what's coming in the future with robots and sophisticated computers... that's going to be smarter than us, and God help us. At least let's I hope nobody builds them to look like the things in these pictures. 😳
@@redlipstickmafia Don't be fooled. That *isn't* going to happen anytime soon. They have no idea how it could even work. (I've been in tech for nearly 50 years.) The recent "open letter" from all the big players warning about "the dangers of real AI" was just a distraction. It was "look over there and worry about a thing that doesn't exist, but don't look at what we're doing right now".
It is actual artificial intelligence. I mean the standards for AI and actual human intelligence is very far that some simpler algorithms could be considered to be AI. Though I believe there should be some level to how AIs should be classified (like calling them by 1st or 2nd generations or level 1 or 2) Idk, my research into AIs are superficial at most, but even imitative programs are still AI.
@@ategetho because AI only has a general idea of what humans look like. When it has to generate a new face instead of using an existing image, it ends up in Uncanny Valley.
I think this is fun! But i would caution not to take these too seriously for another reason. It seems clear to me that this AI was trained on European history. As far I can see Bernadette did not specify Western European portraiture but she got images of pretty fair skinned folks and ""historical dress""" from mostly western Europe. I wish the AI had generated some portraits from some other parts of the world or cultures. There is nothing wrong with European history, but I would love to see some more varied cultural dress when searching like this (this goes for search engine results as well)
@@creativeb549 precisely! It would behoove those training these AI programs to broaden the media they are introducing into the system beyond Western Europe.
I wonder if just inputting the search terms in another language would change the results. While search engines can generally translate between searches and results, they do tend to give you more results in the language you searched for. And there's probably a lot more English sources about white Europeans than, say, Egyptian fashion. Try searching in Arabic and see what you get. So the issue goes beyond machines, and back to general western-eurocentrism, sinces machines can only be as "smart" as the humans that build them (or supply data). But yeah! I wonder if one day we can "teach" machines the social situations and implications, like Bernatte also mentions in the video, and if machines then would make an effort to actively consider those factors and try to even them out.
Look up movie stars from the 1940's movies like the Philadelphia story,the movie the Women, I mean this is not the 1960s ,Google updos from 1940 ,it's glaringly obvious and not hard to spot, 1960' was beehive high teased hair
The 1630s men gave off "men who dressed themselves for the first time and did their best" vibes. I am loving this weird and wonderful path Bernadette is going down, with severed heads, game shows, and all.
The more AI art I see the more I understand Ursula Vernon and her confidence that AI cannot yet replace a human artist. Imitation of art thrown in a blender with the top left off and press frappe. See what sticks to the ceiling.😅
I disagree. I was writing up family recipes, with the idea of a small run of books, for friends and family. I'd started making the dishes, photographing them. That's a huge amount of work. So with AI, I switched. Please generate me a water colour of "describe the dish". I want a white background etc. You get the gist. You tweak the results and there you are. Picture generated, cut and paste, job done. It gives a far more pleasing result. It would have taken me over a year's work to paint the same. Somethings don't work. For example paint me a picture of a roast goose, mixed feathers and roast, with odd anatomy. The same way AI has problems with fingers, and I hasten to add, lots of other artists too. But there are solutions to that. Teach the AI from 3d models for example.
This is raw unguided AI though. A model untrained for the specific task at hand and without any sort of guidance through a control net. You can fix nearly all of these problems by putting a bit more work into the AI beyond writing a short prompt.
I think same, but... We already living in capitalism society, where mass production has extremely low cost and quality. I can see a future, where designers and illustrators partially replaced by AI. Of course, results will be bad, but that might generate even more money for people in charge of that all
Such a fun video! An interesting aspect of these generative "AI" programs is that they aren't trained to understand the context of what they're producing. They're "learning" from the examples fed into them but only for the purpose of reproducing an amalgamation of examples with tagged keywords, and those keywords are very important to why things likely didn't turn out well (for the "AI"). Many of the sources used for training probably didn't have a year stamp in the way you attempted to use it, they probably had many training images with vague tags like "17th century" or "Victorian", and whether the learning algorithm actually learned what years are in each century or when certain historical periods occured is impossible to know because of the black box nature of these generative "AI" programs. They don't even know what faces are in any deeper understanding, they probably just know them as the smudgy skin coloured things above the skinny bit (neck) above the broad straighter bit (torso) above the split thinner sticks (legs) that sometimes bend in the middle (knees). Whether it even knows the nuance of portrait (as in portraiture) vs portrait (as in portrait vs landscape) is also impossible to know. While they're constantly improving, they won't grow in knowledge. They'll just figure out eventually that human fingers don't fuse together when hands are interlocked, but they'll never know how they peed in those gigantic dresses 😊
They’re learning from the collective inter webs. The most pervasive part of the inter webs is CP. You don’t even want to know the results you get from asking AI for anything to do with children or children and pets. AI is FULLY cursed by the cursed humans of our society!😢😢😢
Not only that, but because of how the generation is done (effectively squinting at noise and asking how to add detail to it, if I can anthropomorphise it), it will mistake similar looking things hence the "taken the waist girdle and put it up as a necklace". It associates that imagery with "1560", but can't know where it's meant to go. I think in the future image generation models will have more context for imagery and I can see the start of it happening. I'm not really a fan of AI art because of both how the training images were aquired with all the problems of not getting consent, using private medical images, not paying for usage of images, and the general state of the community which just seems to hate and mock artists in general while often stealing their work, but the technology behind it is very interesting and I hope people learn more about how these models work.
I had a suspicion that the portraits would get more accurate the more modern the time period became- the more images that are available in the training database, the more patterns the machine can recognize and reproduce. We simply have less portraits from the 1500s than we do from the 1800s, so I had a theory that once we got past the wide adoption of the printing press and photography, the AI images would become more accurate because there'd be more references to train on. It's crazy how we have hundreds if not thousands of portraits from the 1500s in Europe (not to mention surviving garments) that would allow historians to make confident guesses about what things should look like, but a thousand different images of X is a very very tiny drop in the bucket of the billions of images that comprises its training database, so the AI can't (yet) do the same competently. The AI doesn't even KNOW what patterns it's recognizing beyond the fact that it appears in other images with the same keyword attached. Regardless, it made sense that the Edwardian portraits were the most passable 😅
Construction is how you can tell an AI image from a real one photographed or painted. If you give an algorithm a string of pixels, it won't know what 'construction' is. Hair fusing into the face/ glasses, or splitting/ merging, window panes that don't make sense, all sorts of structural things that look okay.. from far away or with squinted eyes
Yeah they lack an internal model of the world, which is why we should really call these image generators Diffusion Models or GANs- it lacks an understanding of humans, clothes, fashion (intelligence), instead it just associates certain groups of pixels from its training data with certain prompts. It doesn't first construct the concept of a human in Europe in the 1860s of a certain class, poses the model, and paints in the style of the time, it only have pixels and numbers. Of course to go beyond that means you'll have something approaching an artificial general intelligence
Please make Bernaban Dettner a returning character. I loved the humor of this video. Perfect way to start a Saturday -- Bernadette laughing with us over these images is the best.
I usually don't watch much content revolving around AI generated content, as I feel there are so many very serious discussions to be had around it and don't like to take it too lightly. But I have laughed _so_ _many_ times throughout this video. 🤣 The gameshow skit was perfection.
Just for fun I tried running some of these prompts through MidJourney, and the results were better. It really depends on which AI program that you use.
I think something we should pay particular attention to is the language we use around "AI" art as opposed to human art. Like, here Bernadette sometimes goes "I can see what _they_ were going for", like she does when rating historical costume dramas. This is probably just a slip of the tongue, but also... the thing is, no one was going for anything. There was no human involvement in these pictures. And I think we should be careful not to get that mixed up. Language is the first step to normalisation, after all.
I do agree, but at the same time, I’m only used to talking about art etc as if humans made it. Therefore I’m not sure how I’m supposed to word myself instead.
Im saving this video to watch later, but as someone who loves making ‘historical portraits’ for characters I like and i do research and pick and choose what elements i want to keep, every time I see an AI portrait i laugh because no matter how ‘perfect’ they are, they’re so BLAND. Bland and the face always looks off.
Uncanny Valley is almost inevitable when images are made by something that doesn't really understand what a face is. AI only knows the basic shapes and components of a face, along with odd details like women aren't usually depicted with facial hair.
@@myladycasagrande863 oh for sure. Every time someone shares those x movie by x director meme I always think… you couldn’t just photoshop an actor in? That would have looked more realistic than whatever this is
I like how hair changes colour in the portraits. One of the gentlemen has one white and one brown side to his moustache and one of the ladies has a black plait that turns white when it comes in to contact with her dress.
Idk if the reaction was edited out or if Bernadette was just so distracted by the dress being the chair, but the cursed woman at 15:59 also has the most... _creative_ interpretation of 'hands'. One arm goes into the chair, the other replaces the hand with something like a chunky wooden fan for 'fingers' & then there's an extra set of wrists branching off which have hooves instead of hands. 🤔🙃
Thanks a lot for, at least, mentioning the ethical problems of using AI art Bernadette! For those who doesn't know, tools like Midjourney uses art and images from other creators and authors without consent for generate any art piece. I wouldn't see any problem with this idea used as a tool and i see the potential and the possibilities as, for example, a composition tumbnails generator. However, any of the AI art generator is ethically unsustainable. The people surrounding this phenomena is trying to damage the careers of so many artist on purpose. Not to talk that most of them were created by people who do not believe in "giving credit to the artist". The current "database" of images programs like this use to pick parts and ideas to generate images from, are other people art that was harvested by this tools without consent. What it seems a cute or even meme-y tool is actually a great damage to artist from any kind! (however they are working on the industry or not) Currently, artist are gathering around to fight legally the abuse of this tools by reclaiming our rights of image use. Artist like Carla Ortiz and other great concept artist even brought the subject to Washington so the conditions of AI use in art is fairly regulated and used properly without scraping and scavingin from other people work to make shallow and quick to consume art. I wish more people start hearing artist and so many people that are being damaged by this before using it even to make funny or lighthearted content. I believe that AI use can be interesting and useful, but not at expense of other people hardwork and image rights.
and, on top of that, the thing most people don't realise. It learns, it's getting better and better based on the feedback it gets (aka the pictures people select and say 'yeah this is what i wanted'). Even if it's just for laughs. Or a meme, or a DnD portrait. There's lots of 'AI can never replace' and yet in the past months it has grown and evolved so so much, it is hard to differentiate from real art.
@@lindabrouwer absolutely this!! That's why the take of "haha weird hands" or "this doesn't look good" does not hold up! AI art is improving quite a lot! And now there's good hands and art that could pass for genuine art! The whole point of the debate is not if ai art is pretty or not, the point is that it is not ethical to use and is damaging other artists! If we don't move to conversation beyond "pretty art or not" that would mean that we'll choose AI art just because aesthetics once it seems "beautiful enough" for us, leaving the damage made to the art community ignored.
OMG my friend and I just about died laughing. The in-bred look of these portraits is cursed to be sure. AI definitely needs to take a class in human anatomy.
Love the shenanigans! Enjoyable video and definitely shows AI is still in its hallucinating stage lol. And biting the microphone is always a temptation.
What most people don't realize is that artificial intelligence does not exist yet. No computer as yet has become self aware. What people are calling AII these days is just using a heck of a lot of computers, or computer programs, to make different results.
Yeah they're just massive algorithms (and often not very good ones at that, when you dig much deeper than surface level). But it's a new term so just saying algorithm isn't as ~fancy~
Well, that’s what AI is though. I have a degree in AI and work in the field. AI doesn’t need to be conscious to be AI. The AI you see in movies and science fiction is also known as general artificial intelligence, which is AI that is good at a lot of different tasks, just like humans. The AI we have currently is not general AI, but it is AI. It’s very good at one specific task or a few different tasks, but can’t generalize this to other domains.
Thank you so much Bernadette for a thoroughly enjoyable 22 minutes! You actually made me laugh out loud, not an easy feat these days. Please keep it coming!!!
I love that this isn't just 'wow look how weird this looks' but it's more 'well this is kind of right, and this should really be like this' that we've all come to expect from Bernadette.
Most of your videos have at least one funny, charming moment. This one SENT me. Legitimate tears laughing, and yet another sponsor segment that I absolutely wanted to watch. Brilliant, as ever.
It would be really fun to see you attempt to make one of the cursed AI outfits! I'd be interested in seeing what your takes are on what random shapes the programs apply could be in a practical sense, or even trying to recreate the cursed AI images and make them look less cursed!
This was fun. It's really interesting how different those programs in creating images can be. I've seen real cursed images and then there also were quite decent ones. On another note: I hope Bernadette will show us one day how she's doing her current hairstyle it looks fabulous😍
It would be really interesting to see how the AI responds to historical periods and regions where much less contemporary visual material is available, like ancient Rome, or pre-colonial Native American communities. I suspect it would really highlight how AI tends to copy the biases of its creators/source material.
This is such a brilliant exposition of the strengths and weaknesses of "diffusion" based AI images. I'm seeing this in all manner of areas, and, yes, keep an eye on those hands.
Hi you're my hero! Have you ever heard of Jane Lewson? She was a wealthy widow who lived to be 116 year old lady when she died but she wore the same fancy wardrobe styling for her younger years for 80 years or more which was described beautifully in her obituary, as the description was read aloud I instantly thought I wonder if Bernadette could make it just based on this? She was born in the early 1700s and died in the early 1800s this would be a great challenge!
16:18 that historical dress looks like the one from the 2004 Phantom of the Opera movie! The dress during the ‘Think of Me’ scene!! Maybe the costume designer took inspiration from that portrait? Also AI really needs to learn how to do faces! I think I’ll stick to human generated art for now 😅
I certainly hope the fact that you put full commercial production effort into your sponsor ads means that you are never without sponsors. I enjoy your creativity so much.
I believe the AI you used took you literally on "Tudor" so it searched for Henry VIII and Mary Queen of Scotts and used them as the base for the images.
So I use Google images to help with my art endeavors & I have noticed a sharp up tick of AI generated portraits clogging the search results. Sometimes they are really convincing with only the hands giving it away. Which is infinitely frustrating when u find a dress example that looks interesting only to realize it's AI gibberish not ye olde Haute Couture.
Bernadette, you did it again! The creativity, the green screen, the comedy - with just a touch of horror! (just think of meeting any of these ... portrait personas at night)
Did anyone else think the first 1630 guy being critiqued had a joint in his mouth and not half a white mustache? No? Just me? Okay. 🤣 Thank you, Bernadette! That was a hilarious video and we need more.
That was fantastic! I'm thankful for the AI disclaimer and the video was so much fun. Especially the dating game section - I cried from laughing!! The moderation was just hilarious
Love Bernadette genuinely asking the AI to understand dress construction when it doesn’t even understand hands.
It just doesn't understand anything, because there is absolutely no intelligence behind all that
Some of those “fingers” are going to give me nightmares 😂
What’s interesting is even in human-generated art, hands are usually considered the most difficult to draw/paint, so I am comforted knowing AI can’t master hands either LOL
😂😂😂
Or eyes! 😂
i love how Bernadette's aesthetic has gradually shifted from poised but meme-y victorian millenial to just. accepting that she has the power to do whatever the fuck she wants and nobody will stop her
Not keen myself: too influencer-y, imo. I loved her very natural look.
The Gameshow segment was peak snark and beautifully executed. Had me in stitches. Pun intended.
Omg I know I was giddy through that whole segment lol 😂
Agreed! Fabulous
That orange dress tho!!! 🧡
The hair too!
Bernaban Dettner is PEAK
Mildly disappointed that The Dating Game didn't end with you having to choose a very cursed individual to have a candlelight dinner with, but overall this video was hysterical and made my day.
Ikr? Before the game began, I legitimately thought I was going to watch Ms. Banner decide which cursed man to go on a date with🤣🤣
@@kellymun6757 I'd go with the one wearing the chair
I just realised how these AI generated images could potentially confuse things in the future when it comes to historical research, if any of these images get saved and misplaced and miscategorised by people not knowing better or just by mistake. It likely wont happen any time soon, except perhaps in armature circles, but as time goes on if care isn't taken then I wonder if the odds of this getting mixed up could grow.
There is a lot of bad stuff that can happen and the better they get the worse it will be. Anyone could make a picture of you kicking a dog, or of you naked (that is already happening), or any sort of fake news photographs that could sow misinformation.
That’s a very good and scary thought. The things ai could be used to forge or fabricate in the near future could be scary.
I don't think so. I mean, people continue to paint knock-off "historical" portraits (of celebrities, for instance) and I'm not aware of any of those ever getting confused for an actual historical piece. And researchers in the future, with AI being more widespread by an order of magnitude presumably, will be hyper-aware of that potential pitfall.
I think the larger issue is going to be with intentional falsification of history rather than with the accidental kind.
In school you’re taught proper research practices, these kind of things would pop up on google image search or Wikipedia, not on the Smithsonian website or other credible sources
AI lies to us, but subtly, eventually we will believe it.
(Got to keep that discernment in the forefront.)
“Bernaban Dettner” is officially who i strive to be in life (btw that orange dress is AMAZING 🤣)
I know! I want to join the Bernaban Dettner fan club.
She gives me kindergarten teacher sweetness. You just know she's a good person.
I love that Bernaban Suggests that she is not infact human
@@lelalu101Maybe she’s a Time Lady? A cousin of Missy perhaps?
The thing with AI is it's not actually "artificial intelligence", it's just imitative programs. It's of course going to pull some elements of historical accuracy like silhouette, textile details, etc because it is directly copying from historical portraits from the eras you are specifying. The problem is it's not actually "intelligence", and it has no ability to actually understand what it is copying and why, so that's why you get details like a waist girdle being turned into a necklace.
if it were directly copying it'd be way better at it than what this is showing. What its actually doing is trying to come up with its own map of context based entirely on the massive amount of images its been given. it takes a random mix of pixels and then runs it back through its own image identifying algorithm searching for which change in the pixels brings it closer to pinging as the prompt query. So its moving pixels around until it gets a vague "historical portrait of a woman in the 1570s" vibe off what its made. if it were just cutting bits of actual portraits and shuffling them around, you'd expect ruffs that look like ruffs not vaguely approximately ruff suggesting pixels
Thank you for this comment. It helps me a bagel my mind a bit, but my mind is still a bit boggled.😂 this AI stuff is so trippy
Well, this type of AI program is "intelligence." But what's coming in the future with robots and sophisticated computers... that's going to be smarter than us, and God help us. At least let's I hope nobody builds them to look like the things in these pictures. 😳
@@redlipstickmafia Don't be fooled. That *isn't* going to happen anytime soon. They have no idea how it could even work. (I've been in tech for nearly 50 years.)
The recent "open letter" from all the big players warning about "the dangers of real AI" was just a distraction. It was "look over there and worry about a thing that doesn't exist, but don't look at what we're doing right now".
It is actual artificial intelligence. I mean the standards for AI and actual human intelligence is very far that some simpler algorithms could be considered to be AI. Though I believe there should be some level to how AIs should be classified (like calling them by 1st or 2nd generations or level 1 or 2)
Idk, my research into AIs are superficial at most, but even imitative programs are still AI.
Bernadette's excitement for these cursed portraits is SENDING me rn
The faces are so cursed, I couldn’t for the life of me focus on the educational/dress history aspect of this 😅
I think that one from 1760 with the short black skirt should be the star of a horror movie. A truly haunted doll, indeed!
But for real, why are all their faces so Haunting!???!
Agree 💯
@@ategetho because AI only has a general idea of what humans look like. When it has to generate a new face instead of using an existing image, it ends up in Uncanny Valley.
The eyes😳👀👁️
I think this is fun! But i would caution not to take these too seriously for another reason. It seems clear to me that this AI was trained on European history. As far I can see Bernadette did not specify Western European portraiture but she got images of pretty fair skinned folks and ""historical dress""" from mostly western Europe. I wish the AI had generated some portraits from some other parts of the world or cultures. There is nothing wrong with European history, but I would love to see some more varied cultural dress when searching like this (this goes for search engine results as well)
It did that because the data used to train it didnt have other types
@@creativeb549 precisely! It would behoove those training these AI programs to broaden the media they are introducing into the system beyond Western Europe.
@@creativeb549 And it didn’t have that data due to human bias :/
I wonder if just inputting the search terms in another language would change the results. While search engines can generally translate between searches and results, they do tend to give you more results in the language you searched for. And there's probably a lot more English sources about white Europeans than, say, Egyptian fashion. Try searching in Arabic and see what you get. So the issue goes beyond machines, and back to general western-eurocentrism, sinces machines can only be as "smart" as the humans that build them (or supply data). But yeah! I wonder if one day we can "teach" machines the social situations and implications, like Bernatte also mentions in the video, and if machines then would make an effort to actively consider those factors and try to even them out.
Well if you are asking in English you are mostly influenced by English history and history written in English (mostly by English and Americans).
“Wait, she is *wearing* the chair…” i have never been more hyped in my life WHAHAHAHA
Well, there is a patent for a chair bustle, but that was much later. 😆
Your orange dress host costume is a complete success. You look amazing in it in a still Bernadette-all-class-and-style way! This whole thing is fun.
Definitely rocking the 60's. It only needed cat-eye liner!
@A Dedow I think the "Dating Game" was more early 70's.....
Just "Googled" it ! First started 1965, canceled in 1973. Then rebranded for a year as "The New Dating Game".
I was NOT prepared for Bernaban Dettner in all her mod glory. Your 1960s swirled updo brought me such joy and I need the tutorial!
I felt like her vibe was the 40's here
@@brucecampo6543 The graphics and that dress were definitely 1960s
@@ThatWeirdo04 definitely not
Look up movie stars from the 1940's movies like the Philadelphia story,the movie the Women, I mean this is not the 1960s ,Google updos from 1940 ,it's glaringly obvious and not hard to spot, 1960' was beehive high teased hair
@@brucecampo6543 I’ve never seen a dress from the 1940s that looks like that
That Bernaban Dettner lady looks very familiar. Please let her know that we loved the special segment and would love to see her again in the future.
The 1630s men gave off "men who dressed themselves for the first time and did their best" vibes. I am loving this weird and wonderful path Bernadette is going down, with severed heads, game shows, and all.
I hope The Dating Game becomes a feature! Absolutely engaging and the orange dress ?!?!? Too good
The more AI art I see the more I understand Ursula Vernon and her confidence that AI cannot yet replace a human artist. Imitation of art thrown in a blender with the top left off and press frappe. See what sticks to the ceiling.😅
Even if it makes phenomenal art, who cares? A human didn't make it, and it is therefore devoid of soul and worthless.
I disagree. I was writing up family recipes, with the idea of a small run of books, for friends and family. I'd started making the dishes, photographing them. That's a huge amount of work.
So with AI, I switched. Please generate me a water colour of "describe the dish". I want a white background etc. You get the gist.
You tweak the results and there you are. Picture generated, cut and paste, job done.
It gives a far more pleasing result. It would have taken me over a year's work to paint the same.
Somethings don't work. For example paint me a picture of a roast goose, mixed feathers and roast, with odd anatomy. The same way AI has problems with fingers, and I hasten to add, lots of other artists too. But there are solutions to that. Teach the AI from 3d models for example.
This is raw unguided AI though. A model untrained for the specific task at hand and without any sort of guidance through a control net. You can fix nearly all of these problems by putting a bit more work into the AI beyond writing a short prompt.
I think same, but... We already living in capitalism society, where mass production has extremely low cost and quality. I can see a future, where designers and illustrators partially replaced by AI. Of course, results will be bad, but that might generate even more money for people in charge of that all
@@bewilderbeastie8899I wholeheartedly agree
The most relatable AI thing is when it makes sure to hide the demon hands in pockets. Me too, robot. Me too.
Such a fun video! An interesting aspect of these generative "AI" programs is that they aren't trained to understand the context of what they're producing. They're "learning" from the examples fed into them but only for the purpose of reproducing an amalgamation of examples with tagged keywords, and those keywords are very important to why things likely didn't turn out well (for the "AI").
Many of the sources used for training probably didn't have a year stamp in the way you attempted to use it, they probably had many training images with vague tags like "17th century" or "Victorian", and whether the learning algorithm actually learned what years are in each century or when certain historical periods occured is impossible to know because of the black box nature of these generative "AI" programs. They don't even know what faces are in any deeper understanding, they probably just know them as the smudgy skin coloured things above the skinny bit (neck) above the broad straighter bit (torso) above the split thinner sticks (legs) that sometimes bend in the middle (knees). Whether it even knows the nuance of portrait (as in portraiture) vs portrait (as in portrait vs landscape) is also impossible to know. While they're constantly improving, they won't grow in knowledge. They'll just figure out eventually that human fingers don't fuse together when hands are interlocked, but they'll never know how they peed in those gigantic dresses 😊
Cool, had fun reading this.
They’re learning from the collective inter webs. The most pervasive part of the inter webs is CP. You don’t even want to know the results you get from asking AI for anything to do with children or children and pets. AI is FULLY cursed by the cursed humans of our society!😢😢😢
Yes, they will also learn that chairs weren't usually part of the clothing.
Hey, ChatGPT, how did they pee in those gigantic dresses?
*thinking*
According to Bernadette Banner...
Not only that, but because of how the generation is done (effectively squinting at noise and asking how to add detail to it, if I can anthropomorphise it), it will mistake similar looking things hence the "taken the waist girdle and put it up as a necklace". It associates that imagery with "1560", but can't know where it's meant to go.
I think in the future image generation models will have more context for imagery and I can see the start of it happening.
I'm not really a fan of AI art because of both how the training images were aquired with all the problems of not getting consent, using private medical images, not paying for usage of images, and the general state of the community which just seems to hate and mock artists in general while often stealing their work, but the technology behind it is very interesting and I hope people learn more about how these models work.
Maybe a time traveling AI is responsible for all of the cursed medieval cat paintings.
Excellent episode.
And all those creepy babies
This is the best possible explanation. 😂
I had a suspicion that the portraits would get more accurate the more modern the time period became- the more images that are available in the training database, the more patterns the machine can recognize and reproduce. We simply have less portraits from the 1500s than we do from the 1800s, so I had a theory that once we got past the wide adoption of the printing press and photography, the AI images would become more accurate because there'd be more references to train on. It's crazy how we have hundreds if not thousands of portraits from the 1500s in Europe (not to mention surviving garments) that would allow historians to make confident guesses about what things should look like, but a thousand different images of X is a very very tiny drop in the bucket of the billions of images that comprises its training database, so the AI can't (yet) do the same competently. The AI doesn't even KNOW what patterns it's recognizing beyond the fact that it appears in other images with the same keyword attached. Regardless, it made sense that the Edwardian portraits were the most passable 😅
Construction is how you can tell an AI image from a real one photographed or painted. If you give an algorithm a string of pixels, it won't know what 'construction' is. Hair fusing into the face/ glasses, or splitting/ merging, window panes that don't make sense, all sorts of structural things that look okay.. from far away or with squinted eyes
The best is when the thumbnail looks perfect, but then you click on it and the face and fingers are a mess
Yeah they lack an internal model of the world, which is why we should really call these image generators Diffusion Models or GANs- it lacks an understanding of humans, clothes, fashion (intelligence), instead it just associates certain groups of pixels from its training data with certain prompts. It doesn't first construct the concept of a human in Europe in the 1860s of a certain class, poses the model, and paints in the style of the time, it only have pixels and numbers.
Of course to go beyond that means you'll have something approaching an artificial general intelligence
Please make Bernaban Dettner a returning character. I loved the humor of this video. Perfect way to start a Saturday -- Bernadette laughing with us over these images is the best.
Perhaps in a sponsor segment?
I usually don't watch much content revolving around AI generated content, as I feel there are so many very serious discussions to be had around it and don't like to take it too lightly. But I have laughed _so_ _many_ times throughout this video. 🤣 The gameshow skit was perfection.
We need a "sew the costume produced by AI" challenge.
Just for fun I tried running some of these prompts through MidJourney, and the results were better. It really depends on which AI program that you use.
#I appreciate Bernadettes expertise but prefer kahmas lively snapshots.
"I made this CURSED AI HISTORICAL COSTUMES into reality" next series maybe? i would looove to see that from you!
Bernadette hosting her own game show is something I didn't know I needed 😂😂😂. I think this might be my favorite video ever
That orange dress just screams 60s glamour and I NEED it in my life
I admire your ability to entirely skip over the horror of the faces on every one of these.
I ADORE Bernaban Dettner!
THE RETRO GAMESHOW SEGMENT Bernadette you absolute legend
I think something we should pay particular attention to is the language we use around "AI" art as opposed to human art.
Like, here Bernadette sometimes goes "I can see what _they_ were going for", like she does when rating historical costume dramas. This is probably just a slip of the tongue, but also... the thing is, no one was going for anything. There was no human involvement in these pictures. And I think we should be careful not to get that mixed up. Language is the first step to normalisation, after all.
As an artist who's very irate about AI right now, this is an important point.
I do agree, but at the same time, I’m only used to talking about art etc as if humans made it. Therefore I’m not sure how I’m supposed to word myself instead.
Ya'll are exhausting.
@@SuvuIC maybe "I can see what happened here"? Or "I can see where this came from", referencing the training data?
Im saving this video to watch later, but as someone who loves making ‘historical portraits’ for characters I like and i do research and pick and choose what elements i want to keep, every time I see an AI portrait i laugh because no matter how ‘perfect’ they are, they’re so BLAND. Bland and the face always looks off.
Uncanny Valley is almost inevitable when images are made by something that doesn't really understand what a face is. AI only knows the basic shapes and components of a face, along with odd details like women aren't usually depicted with facial hair.
@@myladycasagrande863 oh for sure. Every time someone shares those x movie by x director meme I always think… you couldn’t just photoshop an actor in? That would have looked more realistic than whatever this is
This stuff isn't artificial intelligence - it's just a plagiarism tool. It just cobbles together images and text that people made.
Hysterical. The faces were a riot. Thank you for very interesting look into AI as applied to fashion.
The wearing-the-chair one is giving me some really horrifying almost Tim Burton sort of feels with a nice dash of uncanny valley mixed in
They all look 'influenced' by Salvador Dali.... Love that orange dress on you! Wonderful video as always. 👍
the production value with the game show in-between was just so high! it was beautiful.
I like how hair changes colour in the portraits. One of the gentlemen has one white and one brown side to his moustache and one of the ladies has a black plait that turns white when it comes in to contact with her dress.
Idk if the reaction was edited out or if Bernadette was just so distracted by the dress being the chair, but the cursed woman at 15:59 also has the most... _creative_ interpretation of 'hands'. One arm goes into the chair, the other replaces the hand with something like a chunky wooden fan for 'fingers' & then there's an extra set of wrists branching off which have hooves instead of hands. 🤔🙃
Oh good, I totally came here to see if anyone else thought that looked like a hoof! It was all I could look at. 🤣🤣🤣
Omg i started freaking out in the sponsor when i saw your anne Boleyn necklace i have one and i love it so much
I hope we see more Bernaban in the future! I love her smarmy contrast to your gentle silliness.
The floating ruff made me think about what tudor wizarding fashion might have looked like. It would be unmistakenly gorgeous :D
I loved “The Dating Game” 😂 Bernadette’s humor is top tier 🏆
Thanks a lot for, at least, mentioning the ethical problems of using AI art Bernadette!
For those who doesn't know, tools like Midjourney uses art and images from other creators and authors without consent for generate any art piece. I wouldn't see any problem with this idea used as a tool and i see the potential and the possibilities as, for example, a composition tumbnails generator. However, any of the AI art generator is ethically unsustainable. The people surrounding this phenomena is trying to damage the careers of so many artist on purpose. Not to talk that most of them were created by people who do not believe in "giving credit to the artist". The current "database" of images programs like this use to pick parts and ideas to generate images from, are other people art that was harvested by this tools without consent.
What it seems a cute or even meme-y tool is actually a great damage to artist from any kind! (however they are working on the industry or not)
Currently, artist are gathering around to fight legally the abuse of this tools by reclaiming our rights of image use. Artist like Carla Ortiz and other great concept artist even brought the subject to Washington so the conditions of AI use in art is fairly regulated and used properly without scraping and scavingin from other people work to make shallow and quick to consume art.
I wish more people start hearing artist and so many people that are being damaged by this before using it even to make funny or lighthearted content.
I believe that AI use can be interesting and useful, but not at expense of other people hardwork and image rights.
and, on top of that, the thing most people don't realise. It learns, it's getting better and better based on the feedback it gets (aka the pictures people select and say 'yeah this is what i wanted'). Even if it's just for laughs. Or a meme, or a DnD portrait. There's lots of 'AI can never replace' and yet in the past months it has grown and evolved so so much, it is hard to differentiate from real art.
@@lindabrouwer absolutely this!!
That's why the take of "haha weird hands" or "this doesn't look good" does not hold up!
AI art is improving quite a lot! And now there's good hands and art that could pass for genuine art!
The whole point of the debate is not if ai art is pretty or not, the point is that it is not ethical to use and is damaging other artists!
If we don't move to conversation beyond "pretty art or not" that would mean that we'll choose AI art just because aesthetics once it seems "beautiful enough" for us, leaving the damage made to the art community ignored.
OMG my friend and I just about died laughing. The in-bred look of these portraits is cursed to be sure. AI definitely needs to take a class in human anatomy.
Love the shenanigans! Enjoyable video and definitely shows AI is still in its hallucinating stage lol. And biting the microphone is always a temptation.
Your sense of humor is delicious, wagging and engaging. Am a fan almost ab initio of your channel. Fascinating to watch you build your world. Bravo.
What most people don't realize is that artificial intelligence does not exist yet. No computer as yet has become self aware. What people are calling AII these days is just using a heck of a lot of computers, or computer programs, to make different results.
Yeah, the AI wasn't "trying" to do anything. It wasn't "getting at" anything. It has no will of its own.
Yeah they're just massive algorithms (and often not very good ones at that, when you dig much deeper than surface level). But it's a new term so just saying algorithm isn't as ~fancy~
Well, that’s what AI is though. I have a degree in AI and work in the field. AI doesn’t need to be conscious to be AI. The AI you see in movies and science fiction is also known as general artificial intelligence, which is AI that is good at a lot of different tasks, just like humans. The AI we have currently is not general AI, but it is AI. It’s very good at one specific task or a few different tasks, but can’t generalize this to other domains.
Does the AI know it’s an AI?
Yes it’s just photobashing basically. From real artists, which is why they’re having lawsuits rn.
Had AI been an actual portrait painter, I'm not sure they would get any further commissions.
Bernaban Dettener is my new favorite thing.
Thank you so much Bernadette for a thoroughly enjoyable 22 minutes! You actually made me laugh out loud, not an easy feat these days. Please keep it coming!!!
As a musician, I appreciate your music choice of 'The Web of Lies'
I realize this segment was focused on AI generated period clothing, but the physical elements of the portrait subjects are hysterical!
She's rockin' that orange dress!
I love that this isn't just 'wow look how weird this looks' but it's more 'well this is kind of right, and this should really be like this' that we've all come to expect from Bernadette.
Most of your videos have at least one funny, charming moment. This one SENT me. Legitimate tears laughing, and yet another sponsor segment that I absolutely wanted to watch. Brilliant, as ever.
It would be really fun to see you attempt to make one of the cursed AI outfits! I'd be interested in seeing what your takes are on what random shapes the programs apply could be in a practical sense, or even trying to recreate the cursed AI images and make them look less cursed!
I love the editing on this video! It was still very interesting and I learned a lot, but the little game show was perfect 😂❤
This was fun. It's really interesting how different those programs in creating images can be. I've seen real cursed images and then there also were quite decent ones.
On another note: I hope Bernadette will show us one day how she's doing her current hairstyle it looks fabulous😍
Bernadan Dettner is a Treasure! Love the video.
Plot Twist: AI stands for Arduous Incompétence
You just keep getting better and better!
The o n e finger in one of the portraits in the first set is comical JWHAJSJAH
I *love* that woman wearing the chair. Christian Tagliavini could launch an entire series of photographs with that concept.
It would be really interesting to see how the AI responds to historical periods and regions where much less contemporary visual material is available, like ancient Rome, or pre-colonial Native American communities. I suspect it would really highlight how AI tends to copy the biases of its creators/source material.
Great topic. Bernadette make THE BEST sponsorship ad's! I enjoy watching them so much.... with other RUclipsrs I fast forward.
I find it hilarious that, much like real painting, AI also finds that hands and fingers are basically impossible to get right.
yes hands are even hard for ai an hoomin alike
Real painters are actually fairly good at painting hands, don't disparage them because algorithmic theft is shit at understanding anatomy.
@@jasminv8653 I think they meant that it’s difficult to learn. It is difficult to learn, speaking from experience.
@@Lauren_P_ I'm an artist too buddy. :) Learning anatomy just takes some theory and rehearsing as any art fundamental does.
@@jasminv8653 u clearly don’t understand what they’re saying or just don’t want too.. 😂ppl like u annoy me
It feels illegal this content is free. Such quality. Much wow.
The game show dress deserves its own show!
The lady in pink and white looks like the illustrations from the scary stories to tell in the dark series
This is such a brilliant exposition of the strengths and weaknesses of "diffusion" based AI images. I'm seeing this in all manner of areas, and, yes, keep an eye on those hands.
Hi you're my hero! Have you ever heard of Jane Lewson? She was a wealthy widow who lived to be 116 year old lady when she died but she wore the same fancy wardrobe styling for her younger years for 80 years or more which was described beautifully in her obituary, as the description was read aloud I instantly thought I wonder if Bernadette could make it just based on this? She was born in the early 1700s and died in the early 1800s this would be a great challenge!
love your work Bernaban Dettner
The pink / white creation near the end of the video reminds me of the ripped dress in Disney's Cinderella [animated].
Love the orange outfit on gameshow Bernadette,hope we learn about it.
the dating game needs to make a return i was DYING OMG bernabatte danner😭😭😭
Caroline Winkler has a design/ decor & lifestyle channel and she did this and generated milkmaids DJing....
Lmao, I was thinking of Caroline when I clicked on this 😂
Okay, but I'm in love with the game show outfit. I love the color, the cut... everything. It's glorious.
16:18 that historical dress looks like the one from the 2004 Phantom of the Opera movie! The dress during the ‘Think of Me’ scene!! Maybe the costume designer took inspiration from that portrait?
Also AI really needs to learn how to do faces! I think I’ll stick to human generated art for now 😅
Bernadette's use of nondescript blue liquid to represent blood (like every period product commercial ever made has done): 'Tis Peak.
Ok the dating game is amazing, I want this with all the historical costumers on RUclips 🤣
I do believe i will be having nightmares about some of these pictures. The faces were positively frightening!
It looks like Bernaban Dettner is flirting with 1950s style.
I'm so happy to see this video! Patreon messages for the win! 🎉
15:50 (My 14 year old)
“Is that George Washington with boobs?!”
No lie found.
I can't unsee that now, what a brilliant observation
You gotta put a warning on stuff like this. If I'd been drinking I might have spat on my computer. 🤣🤣🤣
@@sarahr8311 lol! Sorry?
I certainly hope the fact that you put full commercial production effort into your sponsor ads means that you are never without sponsors. I enjoy your creativity so much.
I believe the AI you used took you literally on "Tudor" so it searched for Henry VIII and Mary Queen of Scotts and used them as the base for the images.
Yes! I feel like one of the first pictures literally stole thr hands off of a Queen Elizabeth portrait
That's not how it works though.
Regarding the fingers in the 2nd portrait, they remind of the Richard III painting.
we need to have more Bernaban Dettner
That orange dress looks stunning on you!
Want!
So I use Google images to help with my art endeavors & I have noticed a sharp up tick of AI generated portraits clogging the search results. Sometimes they are really convincing with only the hands giving it away. Which is infinitely frustrating when u find a dress example that looks interesting only to realize it's AI gibberish not ye olde Haute Couture.
Yeah, it would be nice if AI generated stuff was tagged that way so you could exclude it from a search if it wasn't wanted.
Bernadette, you did it again! The creativity, the green screen, the comedy - with just a touch of horror! (just think of meeting any of these ... portrait personas at night)
I'm highly against A.I imitation of art in general. Until A.I can create without taking from artists without consent, I will never support it.
The AI nailed the creepy smiles, 10/10 would not hang in my haunted mansion
new bernadette alter ego unlocked: Bernaban Dettener
Did anyone else think the first 1630 guy being critiqued had a joint in his mouth and not half a white mustache? No? Just me? Okay. 🤣
Thank you, Bernadette! That was a hilarious video and we need more.
We're living in a time when people want to scramble history and facts and just about everything else. AI fits perfectly with that goal.
Yep. Misinformation galore. Photo evidence? oop, good luck!
That was fantastic! I'm thankful for the AI disclaimer and the video was so much fun. Especially the dating game section - I cried from laughing!! The moderation was just hilarious