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/
I just wanted to add a bit here as its bothering me quite a lot and I would hate to see a Zazzle ban on Ai images.
There are many different types of creators as far as I can tell here, there are those who purchase stock photography to use in their designs, those who use Canva (who have released their own text to image generator if anyone is interested) then there are those who create their designs from scratch.
I draw, mostly in pencil, I doodle, on scraps of paper, the back of receipts, the shopping list, everywhere. I can take my drawings and alter them in photoshop, or krita or gimp to make my designs, and I can also upload them to ai tools to create variations, to polish them up or just be inspired and get new ideas. In that way ai is very much another tool for artists.
If I want to draw, lets say a particular flower, I may look it up, Im no flower expert, I may look through hundreds of images of that particular flower to get the finer details in my mind. Then I go make dinner, get the kids to bed, have a cup of tea and start drawing that flower.
The flower exists, there are many hundreds of images of it available online, in books etc. Is my drawing infringing on anyone elses drawing of that flower? I am drawing it from my interpertation of all of the images I have seen. Now if I were to take someone elses drawing of that particular flower and copy it, then yeah, I would be, but Im not. I have seen many images of dogs in my life, I have seen dogs, owned dogs. I have a pretty good idea of what a dog looks like. I dont have to take parts of other peoples images of dogs and glue them together to create a dog.
This is what training means, it means looking at many different examples of a thing and creating an idea of that thing in your mind so when I draw it, its is a new image pulled from my memory of what it is.
Ai training is similar to an artist who spends maybe years looking at female portraits, in art galleries, at people on the street, magazines and so on. When they create their own female portraits, they are inspired by all of it but they are not copying it.
In terms of Ai, they/it? Are fed millions of images of lets say eyes, yes, indiscriminately, eyes from all over the internet, in every shape and form from millions of different sources, until it can accurately understand that humans have two eyes and spiders have eight. When an ai tool is used to create an image of a person with eyes, they are not selecting one of the millions of images they were trained with and sticking it on the face of another image they were trained with then adding the lips from another image. The concept of what eyes, lips, features should look like has been formed and is created from scratch using the text description it is given.
If you watch a video of the ai creating the image, it starts with a static, white noise image of nothing and builds the image pixel by pixel. Its not a collage of someone elses work any more than my flower is a copy of anyones elses.
Ai is new, its hard to imagine how it does what it does and as someone has mentioned, you could use it to create a copyrighted image of batman because it has viewed thousands of images of batman and has a general idea of what a batman should look like. But that is up to the user, the same as if I were to take an image of batman to photoshop, alter it a bit and post it for sale. I shouldnt be using batman either way.
So if during training, ai looked at 100,000 images of human hands to understand what a hand looks like, and then creates a hand when prompted, whos is it? We already know its not just selecting one of the 100000 to use and clobbering it together with someone elses fingers selected at random from 100000 other images. It knows hands have 5 fingers (although it gets hands wrong all the time) and it created a hand, its not my hand from my facebook profile 5 years ago. Who should be compensated for the hand? Im not saying whos images it should or should not have been trained with, Im just saying it is not copying any particular image it has been trained with.
So when I upload an image I have drawn myself by hand,as a concept of what I want, is it wrong for me to use it because I used ai?
I apologise for the very long reply, im sick today and have been up for hours, its 7am here now and I have promised the kids we will bake our traditional christmas cookies this morning, but this has been on my mind, I am working so hard on my zazzle stores and I would hate to lose it all because I have used ai in the creation process.
Also, you can type something random into ai and get a nice random picture, or you can work on an idea, adding detail, remixing it and spend hours or days on that one concept to get the image you are imagining, the angle you want, the color scheme, the details, the facial features, the lighting, shadows, whatever it is you are trying to create, in that way it takes skill and time and is a great tool for artists to create a concept.
I dont consider myself an artist, im a doodler who uses many different programs to build upon my doodles, and I almost never publish my scribbling anywhere, i rarely show them to anyone in fact but since ive been up all morning and drawing before I typed this, ill add this mornings doodle here because I will most likely use it when im feeling better to created a fuller image with ai and will then edit it in krita, then maybe canva before I put it up on some product on zazzle.
But if the request is for a fairy sitting on a branch and with wings wrapped around it, long blonde hair with a daisy crown, and with a blue dress, then the sources will be far fewer, those sources likely more recognizable by the original artists.
I think you're thinking in terms of how tags & search works on Zazzle in that you'd be looking for something that contain all of the above in one result. But that's not how this works, the AI is not looking for an image that matches everything in that description, flipping it, changing it's colors and then spitting it back out. Instead, it has learned through training what a fairy is, what a branch is, what a crown is, what a daisy is, what blonde hair is .... and it generates a brand new image incorporating all those random things it learned, in a sometimes surprising composition.
Here's an example I made using Night Cafe. My prompt was
"grasshopper wearing a top hat dancing with a dandelion flower in a sunlit field".
If you put this image into Google image search and look at the right pane at "visual matches" none of the approx. 60 images there come close to looking like this image. I can't even find any individual elements that look similar, such as the top hat or a dandelion or the bug.
This model was trained on 2 billion images in the English language subset of LAION-5b. Can someone tell me specifically which image(s) / artist(s) had their copyrights infringed on here in my grasshopper image?
Heavily edited and re-rolled a multitude of times. Dropped into Canva for some extra elements. We didn't quite get the hands perfect- so that has been cropped. Enjoy!
I want to give another example that I hope will also illustrate the principles at play, in addition to the good examples upthread. Upthread others gave examples of written prompts that gave results which had no match in reverse image search. Another thing that can be done is inputting an image of your own and modifying it.
When I went to Turkey I took a photo of the Sea of Marmara from one of the islands. I input this into Midjourney (which always gives 4 iterations) with a modifier of being in the style of William Morris. The base of the Sea of Marmara photo is still there (although it has morphed into a beautified and simplified illustration rather than a photo). But if you know William Morris, you recognize that "add plants and birds" is the William Morrisification of the image. There is no William Morris design that looks quite like that but it's the general concept (and those plants need a fair bit of Photoshop paintover).
My photo + name of artist who is in the public domain= ethically clean image. Obviously there is the potential for abuse, and much of the AI images floating around the web right now would be flagged by Zazzle anyway since trademarked comic book characters and famous actresses in skimpy clothes are common. But it isn't inherently unethical. The current concerns voiced by living artists and celebrities are being addressed in the new release. It's a tool with a capacity for abuse but all tools have that capacity. Porn and gore got addressed before it got released. Now living artists and celebrities and "don't include me in your dataset" image tagging will be addressed.
Adding your own images is a great way to find inspiration for more designs, if your images contain a lot of text, the results will not be useable, midjourney cant do letters well, at least not without a lot of work, my forum sig/logo took a day of reworking just to get the T and P
As for the op's family images, ai seems to have trouble adding more than one or two people or animals into an image, and you can get results like the family ones above, but it gets better at things as you go along, when I started playing around with it all the faces looked awful, it took time and rewording my prompts over and over to get better faces, which is now much easier. Either I learned or ai did or a combination 😂
Hands are still very difficult so a lot of images that are otherwise pretty good become unuseable.
you can see the hands are a mess in this one. Results are better if you use your own image to start with, like my drawing from earlier which wasnt finished and was very roughly an idea of what I wanted, but i ran it quickly through ai and although it still needs work its not bad
Theres a great opportunity to be had for anyone like yourselves who are already creative and experienced, there will be a lot of people trying to sell their work from ai, but not as many who have your experience in how to sell it and what sells and what doesnt
Just for fun I tried. I don't think it knew what I was talking about. There is a clear attempt to have a polar bear hidden behind a snow covered tree in the lower left image. I did say "hidden" after all. And these are snow scenes. I don't see any mittens although there is a knit headband in one. There are some snowmen in the background of lower right image. I don't know what's going on in upper left image. I think it got very confused by my request to have things hidden within the snow scene.
Excellent! Absolutely tried to get things hidden- the only other thing I can think of is to create a knolling of elements and then build- consistency in style and a match to a snowy scene would be tough- not impossible, but tough right now. Betting a few Photoshop experts could run with this.
