Free AI Image Generation: Demos & Dangers

[Music] welcome to another video from explaining computers.com this time we're going to delve into AI image generation specifically we're going to start with demos of stable diffusion online Bing image Creator and Leonardo AI these are some of the best free applications for generating images online and once we got a feel of what they can do I'm going to discuss the broader implications of this new family of AI tools right here we are on the website for stable diffusion online and like all of the applications we're going to look at stable diffusion is a deep learning generative AI model that converts text prompts to images and what this means is that stable diffusion has been trained on a vast data set of captioned images and it draws on this training data to generate an image from scratch when provided with a text description to use the free version of stable diffusion we don't even need to create an account in fact it's the only application we're going to look at we don't need to create an account for rather all we have to do is to scroll down to uh get started now and to click on get started now where we need to enter a prompt and when communicating with an AI prompt engineering is a whole art form in itself but in this video I'm going to test all applications with some pre-written prompts that I've got over here in notepad there they are that's the first one so let's uh paste that one in here like uh that and there were various Styles we can choose in fact there's loads and loads and loads of styles we just go down a little bit here masses and masses and masses of stars but we'll stick with a cinematic default and there are also some Advanced options we can put in if we wish negative prompts things we don't want to appear in the image and we can also control the seed the seed is the random number used to initialize image generation and we can also change the guidance scale which determines how closely the image generation will follow the text prompt but for now we'll get rid of the advanced options just go back to our prompt as we've entered there and we'll click on generate and there it goes we've got a a counter down here but we'll use the magic of film making to speed on through and here we are this should be our tall fairy tale castle made from cheese possibly I don't know it looks a reasonable attempt I guess let's just download the image so I can use it later there we go and uh we'll now just try out some of my other prompts so I'll just put those in this is going to be our pink SP Ider crawling over microprocessor and uh oh that's not bad I quite like that that's quite a successful image of PR spider calling over a microprocessor I will save that as well and we'll now move on to my next prompt challenge very exciting what will it do let's put this one in here like that and generate and uh oh well it's a rabbit it's not blue spotty but it's wearing uh blue and green clothing it is eating carrots with a knife and forks though again that's not too bad again we'll save the image and finally we'll move on to something that should be very exciting cyborg Panda that should actually be a cyborg shouldn't I typed it wrong in here there we are let's try again cyborg Panda balloons in background so let's try that let's put that in here there we are and a generate and uh oh look a cyborg panda with balloons in background again that's not bad I'm quite impressed with stable diffusion online we haven't had to create an account and we have been able to generate some quite impressive images greetings here I am back again and we're now running the Bing image creator from designer strange name for a product but it's from Microsoft they all have strange names and I didn't think this was going to work when I first went in it told me it was too busy to create images but I tried again and as you can see it has worked I put in the tall fairy tale castle made from chees into the prompt and got these images back which uh look rather good don't I particularly like this one that's the one I think I'm going to say that looks rather good there we go but uh let's now go back and try some of the other things we were trying to generate there's the pink spider crawling over microprocessor let's give that one a go and you might notice here there's a little uh counter this is because you get what are called boosts and if you run out of boosts it takes longer to generate images but I've got enough left for now and as we can see it's now done our spiders and these look rather good I particularly like that one actually looks very good that's uh that's very cool isn't it let's download that I can see why this package has got such a good uh rep ation so let's now try my third test there we go blue and green spotted rabbit eating carrots with knife and fork we didn't get a blue and green spotted rabbit in the last test from stable diffusion so let's see how Microsoft does let's just also give us a little bit more space let's do that f11 to give ourselves more space on the screen oh it's quite stylistic isn't it very very different but uh that's rather a nice image although may maybe that one yes I think we'll go with that that's the image I think I will save as a representative here from Bing and finally let's see what Bing can do in generating a cyborg panda with balloons in the background that's a little test isn't it let's give that a go very exciting this one what's going to happen oh those are impressive I like all of those that looks impressive that definitely gets downloaded but uh I might download several of these that's also rather impressive we need more cyborg pandas in the world don't we definitely let's download that too and let's go into another one these are all worthy of being saved I think and we might as well look at this one too this is slightly creepy isn't it I guess a cyborg panda is slightly creepy but we'll save that one as well we'll have a full set of these cyborg pandas right we've now experimented with stable diffusion from stability AI as well as Bing image Creator which gave us free access to the DAR 3 text to image generator from open Ai and which I accessed using a Microsoft account as you may agree both of these allowed us to easily create some decent images however personally I think the results from being image Creator were more impressive particularly when it came to the augmented panda bears but regardless today there are loads of other websites that offer free AI text image generation including playground AI night Cafe crayon Lexica and gencraft meta the makers of Facebook ought to have an offering called imagine although this is not currently available to use in all territories and there are also lots of very impressive paid only services including the legendary mid joury as well as Google's image and 2 however for our last demo before we Ponder the implications of all this let's take a look at Leonardo AI here if we go to launch the app you can see we can log in using an apple Google or Microsoft account or we can create a dedicated account on this platform and as always for security purposes this is what I chose to do it really is a very straightforward process once we're logged in Leonardo AI offers access to a wide range of features for some very impressive functionality here although for this demo we'll stick with image generation where as you can see we need to enter a prompt as in the other systems and we can also select a model to generate our image we can add a negative prompt if we wish although I've not been experimenting with that and over on the left of the screen we also have a number of different controls lots of controls down here one of which is for the number of images generated when we click on the generate button with a particular prompt and this particularly matters because in the free version you get allocated 150 credits per day and you use more credits the more images you generate so I've been generating based upon four images that uses 14 credits per generation if it was two images guess what you'd use seven credits it's also worth noting that all images you create become public for anybody to use unless you turn off the public images control over here but you can only do that if you've got a paid plan and indeed a lot of the controls down here only work on the paid plan this said the functionality available in the free version is very impressive as you can see I've been trying it out already let's just change my scaling here so we can see the results a bit better this was the pink spider calling over microprocessor this is the Cheese Castle let's just take a look at the cheese Castle the interface here already is fantastic for seeing results I'll save some of these later I think that probably I think that one's the best of those isn't it I think that's that's the case and the spiders are rather good pink spider crawling over microprocessor again some good results that's nice that's nice that's nice that's nice that's probably the best but uh I'll have a think about that but anyway let's try our third test prompt which is the blue and green spotted rabbit eating carrots with a knife and fork will this work shall I I'll go back to four I've got just enough credits left to do this in this test so we'll do this and see what results we get what's our rabbit going to look like it's exciting isn't it and uh oh these are quite good aren't they that rabbit has definitely got spots and it's definitely eating carrots with a knife and fork these are very impressive images aren't they oh that one's got some lovely little clo look a little Le tieing things but uh I think probably that's my favorite I think of those that is very impressive I'll certainly download that image and finally we'll try the fourth test the one that's also rather exciting cyborg Panda balloons in background what can it make of that here it goes see what you can do Leonardo Ai and there we are that's a rather Fierce cyborg Panda I guess some of them are quite Fierce aren't they it depends there's various types of cyborg panderas you you may be aware that one's got a lot of balloons hasn't it balloons caning allies back there um I think I think probably number two and having generated some more augmented panda bears I think it's now time to consider the broader implications of this AI technology as we've just seen AI image generation can be amazing however there are still quite a few issues to be aware of and the first of these is surrendering creative control to the whim of the machine because even people become very Adept highly expert at prompt engineering at writing complex descriptions that generate images they haven't got the same level of creative input creative control over that image as a human artist creating something from scratch and of course people who create images by uh typing in words or uh talking to a computer in the future they won't be actually learning and practicing artistic skills and so I do worry we're going to enter a world where we get used to instructing a machine to create an image it creates something and we go oh that's good we're happy with that but it might not be as good or as novel or as exciting artistically as something a human artist would have created secondly there's a image of copyright who owns the copyright in an image created by a generative AI system that's a tricky question and indeed if we look in the FAQ on the stable diffusion online website in answer to the question what is the copyright for using stable diffusion generated images stability AI state that the area of AI generated images and copyright is complex and will vary from jurisdiction to jurisdiction the other way around concerns are being raised about potential Mass copyright infringement by those companies who are actually producing the AI image generation systems because these systems need to be trained and they're trained using data which in practice means millions of images with millions of associated captions because what these systems are doing their neural networks are learning to associate a particular caption a particular description a particular prompt with a certain image so that when we put a a prompt in they can generate something based on that prompt from all the millions of images they they've been trained on but where do these images come from and the answer is normally they come directly from the internet stable diffusion for example was trained using a set of data called Lon 5B produced by a nonprofit company in Germany called Lon and a lot of the images in that data set were just scraped from the internet and therefore you could argue that stable diffusion is making use of the intellectual property of load and load of creators without their knowledge or permission and to that end Getty Images is suing stable diffusion in both the US and the UK alleging that it scraped 12 million of its images and captions without permission to use as training data other image generators such as the D3 model used in being image Creator were also trained on scraped internet data and indeed open AI who make darly 3 have now put in place a system so that creators can opt out their images from training their future image generation models and in one sense this means that they're going to go on stealing intellectual property from websites in a manner that breaks the terms and conditions of many websites until you specifically ask them to stop doing so however the counterargument is that AI models learn from training data the way that human beings learn and so for example as open AI argue after viewing enough pictures of a cat an image model can draw a completely new cat that was not in its training images similar to how a person might learn to draw a cat presumably stability AI will make similar arguments in its defense and it'll be very interesting to see what happens when the gy images lawsuit goes to court probably in 2024 more broadly there are concerns about the impact of AI image generation systems on the creative economy for many years for example I've earned some of my living by licensing the rights to images and video I've created to Publishers and to broadcasters and I've also as a creative professional purchased images from stock libraries and used them in bits of work for clients so I'm used in both directions to both selling works of art to people and and buying things back again and I worry about what happens when we're in a world where people can just type something into a system that's been trained potentially on data that was used without the Creator's permission and it creates images and you don't get this creative economy functioning and that's got two potential impacts one it means people like myself can't earn my living doing that sort of stuff anymore but it also means that there isn't a financial incentive for people to get into the business to actually develop skills artistically to actually produce images and things in the first place and so do we end up with a world where the only stuff we have is stuff where it's created by AIS which have been trained on previous content because no one's producing masses of new content which can be used to to trade the next generation of system that's that's rather sad isn't it and uh I suppose it's a broader issue in the whole economy this that there is a risk that so many people in so many different areas of activity will find that their last job consciously or otherwise is training an AI system anyway what do you think about all this do you really like AI image generation systems or do you worry about the implications or maybe both please let us all know with your lovely human fingers down in the comments section but now that's it for another video if you've enjoyed what you've seen here please press that like button if you haven't subscribed Please Subscribe and I hope to talk to you again very soon [Music]

Channel: ExplainingComputers

Views: 53,552

Keywords: AI image generation, Stable DIffusion, Bing Image Creator, Leonardo AI, dangers of AI, AI copyright, AI, copyright, generative AI, Christopher Barnatt, Barnatt

Id: f2hV-kpG-8g

Length: 18min 11sec (1091 seconds)

Published: Sun Feb 04 2024