New AI Image Generation Program

[Cat]

I read an article about a new AI (artificial intelligence) image generation program called DALL-E that's supposed to be able to generate life like images based on a sentence that you give it. I was all excited thinking that it would be a great way to create some placeholder photos for Christmas cards and the like without needing to worry about model release forms and all that. 

Well... er... um... maybe not! Seriously, I can NOT stop laughing! 

But maybe it would be useful for something else? Anyhow, here's the URL if anybody wants to try it out. https://labs.openai.com/

[ElizabethR]

Cancel 

[Alissanos]

 

it didnt even try to hide my polar bears! Although the one where the polar bear is building a snowman is kinda cute

[ElizabethR]

I see possible snowmen. They are cute. The one in the lower right reminds me of a Coke commercial.

[KeegansCreation]

So cute (but no attempt to hide). I am partial to lower left where the polar bear builds a snow bird.

It seems the AI has no idea what we mean by hiding polar bears, mittens and snowmen within a snow scene.

I think mine looked so weird because I referenced Magic Eye, the name of that technique. So it made...magic eyes?

[Alissanos]

I think I said blended into and concealed or similar, none of which happened, if I had more time id probably spend ages trying to get it 😂

I did spend an embarresing amount of time trying to get not mangled hands one day and another on getting an upside down face which had horrifying results.

Theres a thread in the midjourney discord on difficult prompts and I wasted a lot of time on trying to solve some of them, if you are at all like me dont try to make a grandfather clock lying on its side on the ground, no matter what I did it just wouldnt do it!

[Alissanos]

Or an upside down volcano!

[KeegansCreation]

Haha the infamous hands. And sometimes eyes are odd. I've had to repaint eyes in Photoshop. And like you say, somethings are just impossible (or we don't know the right wording). 

[ColsCreations]

Yeah, quantities and spatial relationships and other intangible concepts are tough for these generators both because there's no consistent shape/form/pixel arrangement for them to learn, and because what they learn is dependent on the alt text that came with the image. Even if it encountered 'hiding' in 10,000 alts, it's an abstract concept rather than an actual thing and could be completely different imagery in every occurrence. There's no consistent visual for it to learn. Same with quantities.

Also, non sequitur, but I wanted a Maine Coon cat drinking coffee the other day and NightCafe wouldn't do it, kept telling me one of my prompt words was forbidden. Other breeds were fine. So some of the prompt filters may be a little over aggressive at this point, not being able to determine yet what is an appropriate use of a word and what isn't.

[ElizabethR]

Try Maine state cat or Maine cat.  I can't tell if you realize which word is the issue. I'm guessing you do. Zazzle's filters allow the word to be used alone in a search and bring up cat products, so I guess it can be used as a tag word on Zazzle without being flagged. I guess Zazzle is more sophisticated than NightCafe in combining words with images.

[Barbara]

Maybe, like Zazzle, NightCafe simply needs to be alerted to the problem, though it's certainly problematic since it's a shortened version of raccoon, not a full-blown, legitimate word. "Norwegian Forest Cat" might produce results since it's very similar in appearance to the Maine Coon.

Colorwash's Home

[Fiorenzo]

These are lovely.

[Barbara]

I'm still watching YouTube videos on AI, some of them showing the strides it has made in a very short period of time. It's still learning, and the more it's used, the more it learns, so it appears we're seeing just the early stages of things yet to come.

Mentioned was that, often, the resulting images are small, and even when you can scale up to larger sizes, get fuzzy.

One of the videos was done by a Zazzler who showed a beautiful image she'd placed on a spiral-bound notebook. I'd link to it if I could remember her name, but even if I could, I'm not sure we're supposed to link to other people's work.

Maybe I'll be able to investigate the brave new world tomorrow.

Colorwash's Home

[KeegansCreation]

They always require post processing. I've had to do paintovers on some. Others I've traced over with the pen tool in Illustrator. The low resolution is an issue that doesn't really come through when posting in this thread. But that can be fixed. I've done fixes of ancient vintage postcards that got scanned so I can deal with this. But people who thought it was going to be prompt and done will be disappointed when they zoom in.

[Barbara]

Heck, every time I've scanned in my own artwork, I've had to retouch like crazy, and the truth is I love doing it. Additionally, like you, @KeegansCreation , I've done a number of vintage pieces that, though they came through quite large (Smithsonian does this), they nevertheless required a lot of work. But it's fun, isn't it?

Colorwash's Home

[Badeesie]

I made a few family Christmas photos at StarryAI last week that look much better than the stuff Cat got last year. There are still some oddities, but are a lot less horrific.

[KeegansCreation]

The level of improvement in just one year is pretty amazing. Last October there is no way you could make AI template photos but now you can.

[SaraMira]

Your ideas are awesome. I'm creating at https://creator.nightcafe.studio/

and pretty satisfied with it. There are issues with hands and some details, but alot depends what prompts do you use and how far can your imagination go.

Here I opened a shop dedicated to AI art: https://www.zazzle.com/store/sangbang

but process of learining to create is long way to go. NightCafe has great community and alot resources are available online to study.

[Cat]

Wow! Thanks for the tip on StarryAI. I gave it a shot and came up with some (remarkably similar) photos to what you've got there. I'm thinking StarryAI is just better than Dall-E.

[KeegansCreation]

In August OpenAI announced that they were preparing DALL-E3 but they didn't give a release date. They need to hurry that up. DALL-E2 has already been surpassed (as shown here) so they need to get DALL-E3 out and it needs to be better than what StarryAI is currently doing.

[Cat]

I'm hoping one of them will eventually figure out how to do hands! 

[KeegansCreation]

😂The hands 😂  Three fingers and a thumb that's too long. Five fingers and no thumb. Maybe your prompt should request six fingers and no thumb and it will screw that up by giving you five four fingers and a thumb.

But it did a good job on the faces. Especially when compared to your post that started this thread almost a year ago with those Halloween ghoul faces.

[Deb]

Just curious.. but  isn't anyone using MidJourney?   (pros/cons? ) 

[KeegansCreation]

I use Midjourney.

Pro: Wide range of artistic styles that are well done, from watercolor to oil paint to photos.

Con: While it is quite good with faces, it still sometimes screws up hands. It also sometimes screws up paws and bird talons, which is really the same sort of error.

Previously DALL-E2 had the drop on Midjourney because it had inpainting and Midjourney didn't.  But Midjourney got inpainting a couple weeks ago so it's now possible to fix those screwed up hands/paws/talons.

[Fiorenzo]

The problem is, you could quickly end up in trouble, if you don't know where they take their reference imagery. I personally wait for Adobe's response. It already works pretty well and the biggest plus is, they will use their own stock imagery, so you're on the safe side, especially when it comes to people. I won't switch to AI-generated imagery, though, but only use it to improve my own works where needed. For mainly three reasons: the quality (for now), the lack of individuality (these images look basically the same and will become mass products), and the competition you will have with tens of thousands of dropshippers selling that interchangeable stuff everywhere on cheapest Chinese products for cheapest prices.

[KeegansCreation]

Breaking News: Adobe Firefly and all Adobe AI just came out of beta today. Starting today, Adobe AI can be used commercially (not allowed while it was in beta, which ended today ). Welp, I'm off to get crazy with their AI. I plan to use it to make mockup backgrounds, for starters. And lets see how it does with faking plausible template photos.

[Fiorenzo]

Thanks for the info. I didn't use their beta yet, waited for things to get smooth and final before risking crashes and whatever. Since I have the PS subscription (the "cheap" photographer one, I don't need all the other stuff but Photoshop), I guess it will be integrated automatically. I'll check it out when available. let's also see how it works in terms of image sizes and licensing prices. I heard somewhat of basic functionality within PS and firefly is separate. Will also check their usage terms, which is extremely important for us.

[sm]

That's what people will look like in future generations. The robots know this and are working on it.

[Connie]

LOL!

[almdrs]

Those things are all stolen images. The AI gets them from the internet, puts them in a "blender", "mix" them and "excrete" them. 

Every image uploaded to the internet is used, like everything is in public domain. Machine learning, or training, means "machine stealing". 

[sm]

Not necessarily. It depends who and what images they are using. Eg, if Shutterstock or Adobe Stock, generate images only from their own stock library and contributors accept terms where they can sell them, then they may be used legally. Or say, if only public domain images or images from sites like Pixabay were used by the AI, then it may also be legal. Using AI that generate images using any images online, could result in an infringement. It's early days and not worth the risk to use AI images on Zazzle products.

The courts will determine whether AIs that use any copyrighted images will get away with generating images using the 'learning' reason, but will they get away with offering them as downloads? Will uses of those blender images get away with selling them commercially? I image that Zazzle has sellers using that AI generated tag in order to quickly delete them if becomes necessary. But removing copyrighted products that generated income does not free anyone of any liability. I may use them commercially in the future, but only after a precedent is set.

A lot of people and businesses are blindly diving into it, even while there are still so many unresolved legal issues. The better the AI image, the more images have likely been put in that blender which probably means those are the riskiest to use at the moment. If I were Zazzle, I wouldn't allow any AI generated images on the site and put a clause in the terms that restricts buyers using them, at least until these legal issues have been ironed out.

It will be interesting to see how it all pans out.

[SaraMira]

That's a misconception .

The algorithm takes your prompts, explores what they mean, and builds something new, layer by layer, rather than extracting pre-existing graphics and dropping them into a new file.

[sm]

You are both correct. They can do it both ways, eg. from Midjourney, "The generator can create art from photos, videos, and other digital sources or scratch"

Even if it creates them from "scratch, it does so because it has learnt from millions/billions of pre-existing images and then uses its knowledge base to create something new. Effectively, it is the same thing. Without all the images that it scraped from the internet, it would not be have the knowledge to be able to come up with a new image.  

[SaraMira]

yes, you can generate image from base photo, there is an option for that. But, for example, if you went to an art gallery to take a look at sunset paintings from 10 different artists, and then you decide to make your own version of sunset painting - is your painting fake just because you saw it elsewhere? Or if you want to be really creative and unique, you paint green sunset and someone tells you there are 3 other painters in arts history who made green sunsets, so your painting is also fake? Its not yours? Maybe ideas are not yours, but art is.

[sm]

If I had to go to an art gallery for inspiration for a painting of a sunset, yes I would consider it fake. Maybe no one else would know but I would know it's fake. Why not just sit and watch a sunset, or 20 of them? I paint and create graphics using my eyes in the natural world, my own photos or using my imagination. 

[SaraMira]

It would be fake if you would recreate the same exact thing. The point is, every artist eventually can be grouped with dozen of other artists based on their style or something they have in common, which doesn't make them fake. Same is with ai. Your imagination came from somewhere, didn't came out of nothing. At least from collection of subconscious impressions. 

[ColsCreations]

Even if it creates them from "scratch, it does so because it has learnt from millions/billions of pre-existing images and then uses its knowledge base to create something new. Effectively, it is the same thing. Without all the images that it scraped from the internet, it would not be have the knowledge to be able to come up with a new image.

That's exactly how it works with humans. We use our pre-existing knowledge base garnered from a lifetime of seeing things to create something new with our own spin on it. Like your caduceus. (Which is great, by the way!)  It's your own personally created artwork but you had to draw on your pre-existing knowledge of what a caduceus generally looks like to create it. Can you go back through your life and determine the one particular image or artist you stole your design from? No, because you didn't steal it, it's your own unique creation based on what you know a caduceus to look like.

Humans, we spend our entire waking life scraping images that add to our knowledge base. The only truly unique work would have to be something created by someone who was totally blind since birth and thus has no pre-existing knowledge base of what things look like, only their own imagination. 

[sm]

My caduceus image is clearly not a true caduceus symbol, which is a symbol dating back to about 4000BC by my Greek ancestors, btw. Every man and his dog has had to have seen it. Mine is an alternative medicine concept which I came up with first in the world before others copied me. Even my hubbies clinic has one as a logo. I contemplate suing them every time I visit but I can't be bothered lol. I did not look at any image when I drew it. I just drew it. I am not a robot and I do not need inspiration from other artists to draw/paint. I have no interest in what other artists do. I am the same with fashion, and began designing and making my own unique cothes when I was 9.

The caduceus image has been around forever. I could have drawn a similar one from memory of seeing them all but instead I drew something completely unique and also because natural therapy is what I have been into since I was young. Most things I create is based on my own experiences, interests, memories, emotions, lifestyle and/or imagination... something a robot will never be able to do. Other things on here are just the dregs from what other people commisioned me to do for them. Most of the stuff on here and on stock, I don't consider as art. They are commercial graphics I whipped up to pass a bit of time. Art is what I sketch or paint by hand. I would never put them on Zazzle or anywhere online.

People are embracing AI but it will eventually replace us all. Stock and POD sites won't need contributors. Why pay out a royalty to a human when a robot can do it all. It won't stop at just image generation. It will create its own templates, tags etc. By using it, we are just helping it improve. Human art is dead and humans are the ones that buried it. Microstock and POD was the beginning of the end for artist. It's all commercialism. 

[Connie]

@smYou are actually proving ColsCreations' point! Sure, your version of the caduceus is your original artwork, but you based it on the caduceus image that "has been around forever." Even if you didn't look at the original when you drew yours, you still at some point in your life, had seen that original, or one similar used in the medical field. If you had never seen that in any way, shape or form, ("learning," in a way) you wouldn't have been able to make your unique version.

I agree that robots can't create real "art," because they lack the human "experiences, interests, memories, emotions, lifestyle and/or imagination"- the "soul" of art, if you will. But I don't think that it is a copyright infringement for AI to "learn from" existing images online, because it's no different than what we do naturally every day of our life.

Also, when I'm thinking about a new design, I go online, study existing similar images, make moodboards, etc. I don't copy what I find, but I see what's already available, what color schemes are used, what works and what doesn't in what's already available, and generally get a "feel" for that type of design.

I don't paint, but it definitely would make sense for a painter to go to an art gallery before attempting to paint a sunset. Sure, you can go watch 20 sunsets, and I'm sure those artists do, but when you go to an art gallery you can see how other artists have translated that real life image, into a painting. How they handled the light and shading, what colors they used to translate what they saw in the real world, how they blend the colors and maybe add a touch of an opposite color that you don't actually "see" in the real sunset, but in the painting it adds just that extra touch of realism. I'm sure all the greatest artists learned from those who came before them, directly or indirectly. EVERYBODY learns from what they've seen and experienced, consciously or unconsciously.

[sm]

There is no point going into this further since you simply don't get that it is not a caduceus. It is mocking one. A caduceus is a symbol with serpents and wings used by conventional medicine. My symbol mocks the conventional medical industry. It is not my version of a caduceus, despite the title having that (it needs to be found and I have so much space in the title to explain it). It is an alternative medicine symbol that mocks a caduceus and it is original, created by me first. The versions you see online are versions of mine. I came up with the concept. This is what commercial designers (and now AI do)... they go around imitating other peoples'/robots' work instead of creating original work. I did my research but instead of looking at images to inspire me to create the billionth version of a caduceus, I studied what NOT to create or to see if this mock has been done before. When I didn't find anything similar, I drew mine. I am not like you. I don't want things everyone else has because original (and decent) is what make the most per image. I don't do trends. I'm a trendsetter, and I have been all my life. While kids were wearing their Madonna look-a-like outfits in school in the the 80s, I was designing and making my own outrageous clothing. Trends are boring to me. I do original or timeless. I look at Zazzle's stuff and it's the same thing over and over and over again. It's boring. If I create something like that (and I have before to try it out because Zazzle pushes it), sure, I may sell because it's trendy. But the buyer may buy mine, or they may buy one of the other thousands of similar ones. That bores me to tears.

This is the awful thing about AI. Soon no artist will ever want to create anything original again because aside from other people copying it, now a robot can do it in seconds. At least with people, original artists can sue them or settle out of court. Now, what do we all do when someone/something copies us? 

And what about AI images that designers will generate? They are not the creator. The robot is the creator but it's likely a creator with no rights to the image because it's generated by potentially billions of other images. So when you generate something and use it commercially, you risk being sued and you risk having your work (and money) legally taken from you. You can create an invitation set today using AI and tomorrow, I can take it all and sell it myself if the law goes that way. But there is no law for it yet so why are people taking such risks? Adobe that says it's using it's own stock to generate images has had heaps of copyrighted AI content in its library. There were Apple computers all over the place with their logo on it.

It doesn't matter what you (or I) 'think' or 'believe' is a copyright infringement because no one knows yet. There's no legal framework for it yet. The people who own the robots don't even know yet. It will all be tested by the courts and eventually written. What happens when some bigshot artist or company sues an AI generator for using an image commercially that copies theirs? You assume the image is such a big mash of heaps of other images, but you don't know what went into creating it. We can call it 'learning' but there are still massive untested copyright issues. I will not sell anything with AI but if the laws say I can and there are no legal implications (which would mean copyright is pretty much over), then and only then will I take all the best AI designs and sell them myself because, one, I can, and two, I have no incentive to create original work anymore. That sound like plan lol. You all generate and I will take since people are okay taking from me. But until the law is laid down, the smartest thing to do is not touch it, at least not commercially. I have a legal background and I have no idea which way this will swing. No one does so I'm sitting tight. 

[Connie]

"There is no point going into this further since you simply don't get that it is not a caduceus. It is mocking one. A caduceus is a symbol with serpents and wings used by conventional medicine. My symbol mocks the conventional medical industry. It is not my version of a caduceus, despite the title having that (it needs to be found and I have so much space in the title to explain it). It is an alternative medicine symbol that mocks a caduceus and it is original, created by me first."

LOL! You can't "mock" something if you never saw the original to be mocked! I'm not saying your version isn't original- I can't judge that because I don't know enough about the origins. But if you had never seen a "real" caduceus, you wouldn't be able to make a caricature! I'm glad you aren't like me- diversity makes life more interesting, and anyway, I have enough competition in my niches.

I have several designs that I originally made because I couldn't find anything similar. But that doesn't mean that I didn't do my research first. How would I know what a lighthouse or a seahorse looked like, if I didn't either go see real ones, or look at pictures on or offline?

I don't believe AI will necessarily replace real artists, unless those artists decide it's easier to generate mass-producted clipart instead of taking the time to hand draw or paint real art. (Which I've already seen on sites like Etsy and Design Bundles, unfortunately).

I don't know what the copyright implications will be, and I really don't care, because I don't like the look of that AI stuff for the most part, and I don't want to ruin my store's aesthetic. But I have found some cute animals and lighthouse clipart that I might use to make patterns for Spoonflower fabric, because fabric loses the watercolor texture anyway. There might be a few high-profile lawsuits because the lawyers like to make money, but I'm SURE the tech big-shots that developed AI wouldn't have done it if they thought they'd be shut down permanently.