AI images just got WAY too real. FLUX 1.1 deep dive

112.99k views5879 WordsCopy TextShare
AI Search
Flux1.1 pro is out! Full review and secret tips for crazy real photos. #ainews #ai #agi #imagegenera...
Video Transcript:
the following images are not real yes all of these are not real these were all generated by the latest and by far the best AI image generator out there right now and that's exactly what we're going to go over today so flux 1. 1 pro has just been released and this is the best image generator out there right now now I'll link to this page which shows some benchmark scores of flux 1. 1 Pro and various Benchmark metrics compared to other leading image generators out there but this is boring stuff so I'll leave this till the end of the video Let's Just Jump Right In and test this out so at the time of this recording there are various platforms where you can use flux 1.
1 Pro first one is called together. a and when you sign up for a free account you actually get $5 in credits which lets you run around a few dozen Generations using flux 1. 1 Pro another plat form is called rubber band.
com but rubber is missing an e anyways I'll link to this in the description below this is another image generator platform and again you can sign up for free and once you do I think you get around like 10 free images with flux 1. 1 Pro another platform is ever art. a this also allows you to use flux 1.
1 Pro but you do need to pay for some credits before you can use it yet another platform is foul. a they have a flux 1. 1 Pro space however it costs around 4 cents per megapixel yet another place you can use it is of course replicates this also costs around 4 cents per image and then you can also use flux 1.
1 Pro on free. com which I'll link to in the description below however you do need to pay for credits in order to use this so anyways I'll link to all of these Platforms in the description below I'm sure a lot more other image generator platforms will provide access to flux 1. 1 Pro in the coming days as well in this video I'm going to use together AI because they give me $5 of credits which allows me to generate a few dozen images so it's pretty simple to use you can adjust the width and the height and that's pretty much it there's no steps there's no CFG scale there's no negative prompt you can only enter in a positive prompt so let's start with something simple like a woman in the city and see what that gives us so here's what we got let me download this now you can see this is a typical flux generation it's clearly AI generated her face kind of looks too polished too perfect to look like a real human and then the background is super blurry there's a lot of depth of field in these images which are typical of flux Generations so this is clearly AI generated and you might be wondering well how do I get these types of images that you showed at the beginning of the video so here's a really cool trick let me just get rid of this prompt and instead we are going to write IMG uncore and any four numbers it doesn't matter it could be like 1018 or 1019 and then CR2 so what this is is it's actually simulating a file that was created with a Canon camera so a photo taken from a Canon camera would have this similar file structure it would have this CR2 format and and so just with this prompt let's see what happens and this is what we get let me download this you can see it's a really amateur photo it's kind of grainy it doesn't have that super blurry background or that super professional feel this looks like just an average photo taken by a regular person let's try a few more so here's another example now let's try changing the number for example so let's try 25 and see what happens and here's another regular photo that looks like it was taken by an amateur human now instead of this you can also add in a simple prompt so for example let's write cat here and see what happens and look at that this is a really normal photo of a cat I mean this does not look AI generated or instead of cat let's write I don't know selfie and see what that gives us and oh my gosh look at that look at this photo it's very grainy and low quality and this does not look AI generated this looks like a real selfie photo taken by someone with their phone how crazy is that now instead of CR2 you can also try heic and then instead of 25 let's set this to I don't know 28 so right now it's image 1028 hi I selfie and let's see what that gives us and here we go here is another pretty amateur looking selfie photo that does not look AI generated you can see the background is not blurry like the flux photos that we've seen before now instead of selfie let's write party and see what that gives us and here we go look at that again this photo looks very grainy and it doesn't look like a professional very po photo that was generated with flux you can see the Shadows here are very sharp I mean this could very well be just a normal photo that's taken from a nightclub I mean we are really at the point where AI photos are becoming indistinguishable from real photos now it's important to note that these types of keywords only work with very simple things like cat or dog or woman let's try like old man and see what happens all right not bad this still looks like an amateur photo however if you give it a more detailed and longer prompt like a woman at a cafe let's see what happens all right so now we're getting that flux AI generated look again where it looks too perfect right the background is super blurry the woman looks too perfect and Polished the colors are just perfect this does not look like a real amateur photo so it seems like this only works by itself or with very short prompts and if you add another longer description or more detailed description it actually lowers or removes the influence of this keep that in mind this doesn't work all the time if you add something afterwards and it's hit or miss so let's try another one let's try four friends at a dinner party and see what happens and here's what we got I mean this looks like a real photo look at the low quality of this photo the graininess this does not look AI generate there's no blurry background this just looks like it was taken with a lowquality phone here's another photo generated with the same prompt I mean this is really starting to blur the lines between Ai and reality now there are a few other tricks I want to show you so how do we get these mirror selfie photos like this first of all if you just input something like a teenage man taking a mirror selfie and let's press generate it's already not bad as you can see from this generation however this looks too good right the background is blurry a phone photo should not look this good it shouldn't have that much depth of field so instead of just a teenage man taking a mirror selfie let's add in some other keywords like lowquality phone photo Snapchat grainy and see what that gives us all right and here is what we got you can see slightly better the background is less blurry and the photo is grainier now for this apple logo I'm not sure if you're taking a mirror selfie should the Apple logo look like this or be reversed let me know in the comments below here's another trick you can try so instead of typing in a description let's try selfie.
JPEG and see what that gives us wo look at that look at how realistic this is It's grainy the background is not blurry this does not look professional this looks like a real selfie photo and note that this is not conventional all right so the best practice for prompting flux and other leading image generators is you need to give it as much detail as possible but with this trick with just typing in the name of the file you get some really cool results like this let's try something else let's try dinner and then I don't know underscore and then let's put a random date let's put 208 112. JPEG and see what that gives us oh my gosh look at this photo again the realism of this is just absolutely insane like if you just saw this photo on social media if someone just showed you this photo without giving you any context at least for me I would never have guessed that this was AI generated but this is the level of realism we have achieved right now can you imagine what flux version 2 would be like this is just wild and I mean you can be very creative you can try different keywords so for example instead of dinner let's let's try something like passport photo and see what that gives us oh my god seriously that's crazy it just generated like a fake ID of someone now as you can see the words are not correct these don't make any sense besides the word passport over here but if you ignore the words holy crap this looks like a legit photo of someone's ID this is absolute Madness I'm telling you you all right instead of passport let's try yearbook photo and see what that gives us I'm having too much fun with this and oh my goodness we have a yearbook photo of someone and this looks like a photo that you know you're taking on a page of a yearbook you can see like to the left and to the right of this photo it's probably another student photo this is just absolute Madness and you know what folks this gave me another idea let's try this out and see what we get I'm going to write the state only fans. JPEG and see what that gives us aren't I a smart guy and this is what we got it's a lady who looks very cold and she's wearing an only fans shirt not exactly what I had in mind let's try again instead of 2022 I think it's too old only fans didn't even exist back then let's try 2022 and see what that gives us oh my God what the hell is this that is creepy as hell man I'm recording this at night right now and this is creepy all right here's another thing to try out instead of only fans let's try verification photo and see what that gives us what the hell what is going on here this is creepy man this the photos I'm generating right now are getting creepier and creepier but let's try [Music] verification photo.
JPEG and see what that gives us whoa okay I have no idea what this is this wasn't really the verification photo I was looking for I wonder what happens if you scan that QR code let me know in the comments if you actually try to scan this and what page it takes you to all right well enough playing around with this jpeg or crr2 orhe I basically this is another technique that you can use and actually this works with flux Dev vers version one as well this is not limited to flux 1. 1 Pro so back to the goal of our video Let's actually test out flux 1. 1 with some harder prompts all right so what I'm going to do now is first of all I'm going to set this to 1024x 1024 and then I'm going to use a series of quite challenging prompts that I've used to test other image generators before and I'm going to show you the generation from flux 1.
1 Pro versus the other top models out there right now which is flux 1 Pro and ideogram version 2 and mid Journey version 6. 1 so you can see for yourself which image generator is the best so first prompt is a woman doing a warrior 1 yoga pose at home let's see what that gives us by the way for your reference a warrior 1 yoga pose looks like this and this is what I got from flux 1. 1 Pro you can see it doesn't really understand what the Warrior One pose looks like and the anatomy is kind of off but again this is a really hard prompt so far none of the image generators I've tested could actually get this except for Google's image gen 3 and Dolly 3 to some extent anyways here are the four best image generators out there side by side with the same prompt so you can see all of them are actually quite similar in terms of realism it looks like mid Journey did a slightly better job however if you look at the hands and fingers of mid Journey they are just God awful and then actually flux 1.
1 Pro and flux one pro look very very similar I I would say there's really no noticeable difference at least for this prompt for flux 1. 1 versus flux one all right next prompt photo of a man giving a TED talk behind him is a big neon sign with the text tedex AI search Spotlight style lighting soft Shadows sharp Focus shallow depth of field let's click generate and see what that gives us and this is what we got so this prompt is quite challenging because it needs to generate text I'm testing if it can actually understand the tedex logo and also add in some additional text and also make this a neon sign and it looks like flux has no problem understanding that the tedex logo should look like this and it got the text correct however it is slightly weird that there's a huge space between Ai and search and his head just fits perfectly in between these two words so that's kind of weird and then oh my goodness here we have six fingers on this hand now note that although flux Pro is one of the best image generators out there it is way better at getting hands and fingers and feat correct you sometimes still get flaws that's just one of the things with generative AI so here are the four image generators side by side both flux and ideogram got the tedex logo correct and got all the text correct mid journey is kind of a mess it cannot generate words and letters correctly plus the tedex logo should not look like this now in terms of just comparing flux 1. 1 Pro with flux 1 Pro again there's really no noticeable difference if flux 1.
1 Pro is better it's just marginally better I mean both of them Follow The Prompt very well there are very few software tools that I use every day but this is one of them thanks to our sponsor turbo type basically you can create custom keyboard shortcuts so that you don't need to keep typing out repetitive things so check this out let's say there's a prompt that I use often in chat GPT let's set the shortcut to/ suum and then I'll type in the prompt that I use often I can even add in a variable to include any text that I copy and then let's say I find an article that I want to summarize well I can just copy the URL and then in chat GPT type in my keyboard shortcut and voila super quick way to write prompts you can also drag and drop shortcuts onto any text field I can also select any text and then right click to easily add it as a shortcut finally it also supports Rich Text for example you can add in bold and italics and add links to your text as well so let's say I need to send out a lot of cold emails with the following template well I can just create a shortcut for that and then whenever I start an email I just need to type in the shortcut and voila the text is already styled and linked for me there are hundreds of pre-existing templates that you can choose from including common prompts for for chat GPT business finance medicine and more this tool saves me so much time every day so definitely check it out and download the free Chrome extension in the link below all right next prompt lowquality Snapchat photo of a Teenage man taking a mirror selfie the photo was shot on a phone and posted in 2015 on Snapchat let's see what that gives us and hey this is actually very good look at the graininess of this photo this indeed looks like a very lowquality selfie photo taken with a phone and this dude just looks like a regular Teenage dude if I were to point out one flaw it's that the lenses on his phone are not really aligned properly but that's pretty much it his hand looks correct everything else in this photo looks pretty damn good so here are the four image generators side by side using the same prompt again quite hard to pick a winner here I actually looks very good you can see it got the lenses on the phone kind of correct at least it looks better than what we have here and then flux one pro is not bad either but I would say I don't know what this stuff on the wall is I would say the worst one here is again mid Journey the dude is not holding his phone up to the mirror to take a selfie so it's not really a mirror selfie but let me know in the comments which one you prefer I would say in this case if you just compare flux 1. 1 Pro to flux 1 Pro this is just my subjective opinion I prefer 1. one's generation over version 1's Generation all right next prompt a teenage woman holding a handwritten note that says verify me 8 2022 2024 lowquality selfie photo let's click generate and see what that gives us and here's what we got so again this is the typical flux look where the background is very blurry and her face looks too perfect and too polished to look like a real human now it got the text mostly correct except for this line here on the F so that's a noticeable flaw and then oh my goodness look at the fingers is this correct I guess well this finger is kind of like splitting into two fingers and then I don't know what this is so the hands are not correct and again this is generative AI I'm showing you this example because even though flux 1.
1 Pro is considered to be the best out there it's still b sometimes gets hands and fingers incorrect so just keep that in mind it's not perfect and then here are the four image generators side by side in this case at least for me I would say the winner is ideogram this looks a lot more like a lowquality selfie photo this woman looks like a real person plus the text is correct the text looks handwritten and everything just looks more realistic both the flux Generations look way too too good you can see with the blurry background it looks like a professional DSLR photo even though we've specified lowquality selfie photo and then in terms of mid journey I actually do like the style of this photo this does look like a lowquality selfie photo however mid Journey just cannot generate text well it's missing the me here there's an extra slash here so yeah I would say in this case at least for me ideogram is the winner flux 1. 1 Pro versus flux 1 Pro again there's not really a noticeable difference in Improvement all right next prompt two hands making a heart symbol let's see what that gives us and this is what we got this looks perfect the hands look very good and they are indeed making a heart shape the background looks again very blurry this is very typical of flux so this is really good for like a stock photo but if you're really going for that amateur lowquality photo then I think this looks too good and here are the four image generators side by side you can see both the flux generations and ideogram just nail it I mean all three of these look very realistic it's hard to point out any flaws with any of these so I think it's a tie I think these three are tied and then mid journey is just a mess it cannot generate hands and fingers but I mean in between these three it's really hard to pick a winner let me know in the comments what you think all right next up is our Infamous prompt a woman showing her palms and SES of feet realistic professional photo depth of field this is a very tricky prompt that very few generators have gotten correct and indeed look at this I mean this looks awful this is not showing her palms and soles of feet in fact let me generate this one more time and see if it's any better yeah you can see this prompt is just really tricky it throws a lot of video generators off so I mean here's what we got there's not much more to say other than it's a fail and here are the four image generators side by side you can see that actually none of these generators actually got the prompt I think ideogram is the closest she is showing her Souls of feet but not her palms and then both the flux Generations are pretty awful mid journey is missing feet altogether so yeah none of these are great I think maybe for flux version 2 it would Ace this prompt but right now none of these could handle this prompt and just a really subtle thing I noticed about flux 1. 1 Pro it seems like the images are slightly higher in brightness and contrast compared to the other images and if you go back to the previous slide again if you look at flux 1.
1 Pro it seems to be higher in brightness and contrast and maybe a bit more saturated too so that's something to note all right next prompt is testing its understanding of context so the prompt is photo of a red sphere on top of a blue cube behind them is a green triangle on the right is a dog on the left is a cat let's see if it can get this and here's what we got and I expected it to Ace this this is indeed a red sphere on top of a blue cube behind them is a green triangle on the right is a dog on the left is a cat nailed it and flux is known to do very well at understanding context and understanding prompts with a lot of detail so you can see comparing the four image generators both the flux Generations are the clear winners I mean ideogram was close but it added an extra blue sphere in the middle so I mean technically it's still correct we didn't say that there can't be a blue sphere but yeah not as good as the flux generations and then mid journey is again just God awful it's missing a cat and is this even a dog and then there's not really a green triangle in the background so yeah there you go and again notice how flux 1. 1 Pro is slightly higher in brightness and contrast and saturation compared to the other images which may or may not be a good thing depending on what you're going for all right now next up another tricky prompt an astronaut riding a giant snail with an iridescent Shell through a desert landscape the astronaut is waving a flag that says I love AI again this is a very tricky prompt that I don't expect a lot of image generators to get correct there's a lot of things in this prompt such as an astronaut riding a giant snail the snail has to have an iridescent shell they have to be in a desert plus the astronaut has to be waving a flag that has the text I love AI not a very easy prompt to get correct let's see what this gives us and here is what we got I would say it's pretty good this is an astronaut riding a giant snail the snail does have an iridescent shell plus the flag does say I love AI it's interesting how it replaces the word love with a heart symbol which I did not really explicitly say to do but whatever let's give it the point anyways and here are the four image generators side by side again you can see flux 1. 1 Pro and flux one pro are very similar but you can see the snail kind of looks better in 1.
1 Pro Plus it follows the text better I would say for this prompt ideogram is the winner here at least to me this looks very good everything looks super detailed and the text is correct the shell looks very iridescent the snail looks a lot more realistic compared to all these three snails so yeah I would have to give it to ideogram speaking of snails I have shown before that flux does not do well with uncommon animals which I'll show you in a second all right next up I want to test its generation of different styles so instead of a realistic photo now I want to get it to generate a watercolor painting of a whale in the sky let's see if it can pull this off and very good this is indeed a watercolor painting of a whale in the sky and the whale does look like a whale it follows the prompt exactly I can't really Point any noticeable flaws with this this is really really good so here are the four image generators side by side for this prompt I would have to say flux 1. 1 Pro is my preferred choice this is the watercolor effect I was going for this does look like a whale whereas for flux 1 Pro it's not really in the sky and then this does not look like a whale I mean this this face just looks weird and then ideogram is not bad but again the whale doesn't look great and it's not really in the sky and then mid Journey it seems like it's just in between these waves again it's not really in the sky like I specified so yeah I would have to say flux 1. 1 Pro is the winner here again notice the slightly higher brightness and saturation of its results compared to the other Generations all right next Let's test out its ability to generate anime so the prompt is an anime girl in the city at night let's click generate and this is what we get pretty good I would say one noticeable flaw is that the characters on the signs are still not real characters and they don't produce real words but none of the image generators right now can actually do that so if you get it to like generate neon signs in a city at night all of them will not get the text correct so there's still room for improvement hopefully version 1.
3 or version two would actually fix this issue other than that I mean this does look like an anime girl in the city at night I would say that that the background looks a bit too realistic for this to be considered anime but this is just the typical blurry background that you get with flux Generations here are the four image generators side by side you can see ideogram the background looks more anime and this looks like it's coming from an anime show or movie whereas the flux Generations look more like I don't know what you call this like anime illustrations or posters and then again flux 1. 1 Pro and flux 1 Pro look very similar I would say 1. 1 has more details but I mean both of them are not bad and then mid journey is just weird this is not anime this is just digital art all right remember when I said that flux sucks at generating uncommon animals so let's test this out so the prompt is a kodo dragon in the forest its tongue is extended and it appears to be alert and in motion realistic photo let's click generate and see what that gives us this is what we got this is definitely not a kodo Drgon which for your information looks like this in real life fun fact this is the largest lizard in the world and you can find it in Indonesia anyways back to this photo you can see it looks nothing like a real kodo dragon but this is expected like I said it's known that flux sucks at generating uncommon animals I was hoping that flux 1.
1 Pro would be better at handling this but no it looks like its training data just did not have these photos of uncommon animals so here are the four image generators side by side and in this case mid journey is the clear winner here this is the only one out of these four that can actually generate a kodo dragon although its tongue is still kind of wrong this should be a forked tongue and not like this but I mean the other three aren't even close this just made up some Drgon likee creature note that if you compare flux 1. 1 Pro with flux one 1. 1 Pro does seem to have more details on the Drgon in fact let me try something real quick and see if this works what if I did komodo dragon and then let's just put a random date 2023 1205 JPEG and see what that gives us just for fun I'm not expecting anything fancy here and this is what we got still not a kodo dragon anyways that sums up my test and comparison of the four best image generators out there let me know in the comments which one you prefer at least for me it seems like flux 1.
1 Pro isn't really like noticeably better there's not a huge difference between flux 1 Pro it's just a really slight difference now going back to this page this is from a third-party independent evaluator called artificial analysis they have this image generator leaderboard and how this elos score works is that it takes in these blind test results from users so users can go in and actually do the blind test in which they compare two image generators side by side using the same prompt and the user basically selects which image generator they prefer now you don't know which image generator is on the left which is on the right and it could be any one of these models now after a ton of users have done these blind tests you can see that flux one pro is preferred the most so it has an Lo score of 1153 a win rate of 69% which is the highest in this table and Then followed by flux one pro followed by ideogram version 2 and surprisingly in fourth place is mid Journey version 6. 1 if you've been following my channel in a previous video I mentioned that there was a mysterious new model on this leaderboard called blueberry which appears to beat everything else well it turns out that flux 1. 1 Pro is the blueberry model so back to this blog they finally revealed that yes they've tested flux 1.
1 Pro under the code name blue in this artificial analys image Arena and they found that it surpassed all the other models on the leaderboard and it achieved the highest Lo score so here are the highlights about flux 1. 1 Pro it provides six times faster generation than flux 1 Pro while also improving image quality prompt adherence and diversity and another great news is they've also updated flux one pro to generate the same output as before but two times faster they've also announced that flux 1. 1 Pro will soon be able to generate ultra high resolution images so you can generate up to 2K images without sacrificing any of the prompt following and what's crazy is the pricing is actually even cheaper than flux 1 Pro so you can see here flux 1.
1 Pro costs 4 cents per image whereas the previous version flux 1 Pro cost 5 cents per image so there really is no reason to continue using flux 1 Pro now do keep in mind that the pro version is closed source so you can't download this and run it locally for that you'll need to either use the dev or the Schnell version but anyways that sums up this release of flux 1. 1 Pro hopefully this video gives you a sense of what it can and cannot do and definitely try out the CR2 or the jpeg promps that I shared near the beginning of the video this allows you to produce really realistic amateur looking photos that are really hard to tell apart from real photos anyways that's a wrap let me know in the comments what you think of all of this what impressive things were you able to generate using flux 1.
Copyright © 2024. Made with ♥ in London by YTScribe.com