Hey brother, what an amazing test you did!!!!!!! Can you tell us what is the best one for creating beards and hairstyles for men and good faces? Please!
For realism yeah, but MidJourney blows all of them away if you want produce art, like paintings, abstract, creative. I think we need to forget about one tool for everything, I really hope MidJourney doesn't screw up it's style chasing photo realism.
I still prefer Flux. The Open Source fine tunes and tools, are still unmatched compared to the Closed Source models. Still this is good. Competition breathes strength and like we've been saying at nauseam since SD1.5 burst into the scene. This is the worst it'll ever get.
Meh flux is actually kinda going openai way. Schnell does not really produced serious production level albeit its fast and can be used for commercial. Dev are not commercially available and pro well its just straight paying them the money. Ideogram actually good but again it has too many limitation for the free tier. Mystic is the same with that other existing one will run out of steam when we used it, unless you got a client who wants to create their own custom platform then yeah not worth it.
@@savire.ergheiz You can use images from Flux Dev commercially, just not the the model or Fine Tunes. They put an enphase of distinguishing the images you generate, from the 'derivatives' (Read Fine Tunes, Loras and Merges). Likely not wanting to repeat the mess that was SD3 and it's Termes of Service.
@@vi6ddarkking Yea lets see later. We have so many examples of companies doing such thing like what Unity did. Its always beginning with simple limit like that. True free open source won't really put any limits albeit it won't do the author any good rather than being famous.
@@savire.ergheiz It's not even clear that models themselves can be copywritten with the current rules. Much less the outputs. So the license is at best enforceable only on those using the model, and perhaps not even them. We'll have to see court cases on this to find out.
This is a great channel… but what we need is the leap from image generator to art directable images using Ai. That will be the true turning point… not what MJ does… but create a video of people walking down the street, then be able to select a tree in the image and say turn it 90 degrees, or make it back lit. Without that the generators are ‘you better like my first attempt, otherwise you’re stuck’. Vfx companies all know this… Sony is working with Sora trying to push this aspect. Hollywood is not going to get affected until this ability goes mainstream. I’m a vfx supervisor for movies, I love Ai… looking forward to the potential it offers but right now, it’s just good for ideas… after that it’s business as usual.
Midjourney FAST & RELAX do not affect quality; as FAST means you get the attention of the server almost immediately; and RELAX means that the server - when less busy - finally picks up your prompt!
Heavily disagree on ideogram being bad at anime. In fact, I like ideogram's Anime style more than the others. Because it looks like actual anime scenes. With the others, I feel like, you can see that it's AI anime, which I don't like! I like the 2d look of ideogram's anime.
Good to see Midjourney being embarrassed with their low-quality crap. Hope this actually wakes them up and face the fact that no one cares about all the social stuff they're working on, instead of the actual model.
The prompts were low quality, so are the expected results. Midjourney outputs the same quality as the others you only have to know how to prompt is correctly.
VPN doesn't help with phishing, because the connection to the email server is encrypted by SSL. If your VPN is able to read your emails and "protect" you from phishing, it means that your data was somehow decrypted by vpn software and compromised. So, your advertisment is nonsense or it was anti-advertisement.
VPN are pretty much useless those days because everything is encrypted at the application level. The only use I see is to switch your IP to a different country to access servers that reject certain IP range, or serve different content.
@@xl000 VPNs are only becoming useless on the security side of things, they are still very relevant for firewall dodging, location spoofing, and anonymity.
We all knowns what VPN's are used for. Sailing the high seas. It always was and as long as you can get cought doing just that, they will be useful. Sure, things like poor journalist is oppressive countries use them also. But I bet 98% of all VPN bandwidth is used for that type of acquired content.
@@ianPedlarTor browser is slow as fuck, so if you want to do stuff thay uses a lot of data like watch videos locked in your country, it's gonna a pain in the ass.
It doesn't matter how good the model is when it starts. What matters is whether you can train it and it can be improved by the community without censorship. So in this respect Flux has a bigger margin to develop itself. The marginal differences between these models is then not important anymore. Flux doesn't limit creation.
Flux is superior just because the fact its open-source and you can bend it in any way and never says no. I wish more people would understand this. There is freedom without censorship. Be it politics, be it image generation or text generation.
At this point, "better" doesn't mean much in terms of quality and adherence. Sure, there are some that are slightly better than others. But what it ultimately comes down to is censorship. I don't need a company telling me what I should be able to create. If do something "bad" with my creation, then I should be punished by the law of the land.
You're not wrong but that's not reality. These things are developed by people, and those people make the rules of what they develop. Not you. So not allowing them to censor their own product is taking away their freedom as well. No one is forcing you to use their product and play by their rules, but you want them to play by your rules.
@@drowzy2309 "So not allowing them to censor their own product is taking away their freedom as well" to me, that makes no sense. Why don't makers of kitchen knives make them all without a sharp point and a bit dull so that they can't be used to stab someone? Answer that and you should be able to answer the AI censorship issue as well. Childproof caps is another one. Good intentions, bad result. Now, for the sake of .1% of irresponsible parents and kneejerk idiots in power in govt, ALL parents, ALL people have to put up with struggling to open a pill bottle. At least now they have invented a DUAL top so you can flip it over and use it as a NORMAL top for those who do not have children. Point is, like Smokey The Bear said way back in the last century, matches don't start forest fires - people do. The tool should not be censored. The people should be punished only AFTER a crime has been committed as defined by each local govt. Companies shouldn't have a say in what other people MIGHT do with an object they made. Once it's created and released into the world, what Joe Blow does with it does NOT make the creators of it culpable in any way just like a match manufacturer isn't responsible for the arson. get it?
I feel like a big advantage Flux has is if there's something that it sucks at (like yoga poses or uncommon animals) you can just run it locally and train a LORA to make it better
For the warrior yoga pose results : there is multiple warrior poses, the one you wanted is known as "warrior 1", while some of the results you got actually matched with other warrior poses, so i wouldn't exactly say that the ai is wrong.
hehehehe how come Pony is never mentioned around Anime... I know its very NSFW but its also very powerful in accuracy and the huge pervy community behind it that are really motivated to also make SFW Lora's
@@skyfe5430 you can but only for a few steps of an entire professionnal worfklow. Clients want specifics images with specifics composition and consistency, it's only achievable using controlnets. Midjourney/ideogram are good for creating nice images on instagram / creating ai video stories, book cover, but for the rest it's not that usable
great comparison, but what you meant was manga, not anime (which are movies) and ideogram made actually the best manga art, also the letters on signs were kind of correct, it just was not english )) it was japanese alphabet katakana / hiragana
The title suggests that the comparison will be made to the maximum capacity of each tool. But this is not the case. When I see the importance of the "Guidance Scale" parameter on Flux Pro, I think you could have had much better results with a value of 2.5 (instead of 3.5). And this is certainly true for the other tools. Unfortunately (or fortunately) to know what a model has in its belly (French expression), you have to invest many hours. I'm still convinced that Flux is the most versatile and has the greatest potential for improvement, but... I'm going to take a look at Ideogram V2. Thank you so much fot this amazing work.
First things first - on several occasions you said the hands on mystic looked good, when the were not. I ran the same prompts on Flux-dev, running locally and I have to admit, I was surprised that it didn't do the hands/feet prompt that well. Also, when removing the "dragon" word from the komodo prompt it did comparably well as midjourny in your case. just the tongue wasn't quite right. All other prompts where either absolutely perfectly following the prompts, or where very close to perfect. I'll definitely try ideogram, as it really stands out in your tests.
Informative video, but the fact is that there are different ways to prompt for different models, and also, there are different tools on each website. For example, I would never even try to prompt for text on Midjourney.
The new Imagen 3 actually does by far the most realistic human anatomy features than any other model out there. It's a lot better than Ideogram 2, Mystic, Flux, etc. Imagen 3 finally got hands and feet right. Literally the human anatomy on Imagen 3 is beyond any of these and nobody is talking about it. That is quite ironic and quite hilarious actually.
@@nemonomen3340 I'm a simple man. I try out various AI image generators and image to video generators and my number one question to all of them in my mind is, "Are you going to let me generate some huge boobs or not?"
I think the tests should be small variations of previous tests, generators could easily incorporate the previous failed prompts in their fine tuning, so cat could be on top if a yellow cube with a red square behind, every time a small variation on the same theme..
Midjourney is the only one that has "character" and "style" references. A great image I can never imagine again makes it good for single use cases only.
A couple of good tests for you: Man eating a banana - MJ doesn't peel it, MS does. Waterskiing - both have ropes all over the place. Which is worse though? Probably MJ.
Number one mid journey has had a website it just hasn't been open for everyone and number two. The speed on mid journey determines how fast you are in the queue to be rendered it does not affect the quality at all. It will actually take the same time to render for all choices.
I think open source is best cause you can train them as you want, You get so much functionality to develop those models. The way you want, You get a control net, an ip adapter, inpaint option , and If you know coding you can make those models the work way you want and can sell those specially built model
i noticed the pattern for midjourney is they use ban on users to basically deny service while keeping your money that you paid to them. its a very strange pattern that i uncovered when discussing with people who paid for annual subscriptions and got banned because of like what you encountered, triggering the NSFW over images that shouldn't really be an issue.
If you test Midjourney anime you should be using Niji 6, not MJ6.1. Niji 6 is a really good anime generator (though it has much less knowledge of IPs than some like NAI). Also, you can get around MJ's moderation by using words like 'president' - because it strongly believes we have only ever had one president, and it was the 45th - but you'll likely end up with him kissing Hillary, rather than Kamala.
for the selfie photo perhaps we should mention which generation phone the selfie was taken 😋 .. the modern phone selfie portrait mode is really good so no surprise flux-pro generated it with all the background blur 😀
One of the better videos on the topic, for sure. Most are biased and/or obnoxious presenters. All in all, I just get more confused every day as the list of image makers grows and each one claims to be The New Best One Ever. The text (prompts) remind me of old text-adventure games-you knew what action you wanted, but the text parser was so limited that it took ages, blindly guessing what verbs and nouns the game had in its vocabulary-any others just got an "I don't understand" response. Today, between text and imagery is a real disconnect between how much "AI" in general can comprehend, much less synthesize.
To this day im so shocked why freepik doesnt get more shoutouts. 100 bucks a year. 200000 credits. Flux, mystic, magnific upscaler. Sketch to AI. RE IMAGINE, INPAINT, EXPAND... PLUS 1 million stock images, templates Plus video stock. Its a bohemoth !!!
The Flux version for the verify me only flaw was how good it was, and I wonder if you took the smaller versions and did the same thing could you tweak the low quality out a bit more. But what really caught my idea on it was how well the sign was done in low quality. The lettering isn't perfect like a font, the f is redrawn over like it was really hand made. Take that photo and run it through a sharpen filter and I think you'd have 100%.
I've been enthusiastic about AI for a while I think is a good thing to have but I can't help but worry as the "why" of all of this? Specially considering how much money is being put onto this. As of now AI models seem to be only good at an extremely specific one thing and in order to reach it they have insane requirements. Even if they come out as a fully realized working machinery I can't help but think that they will be extremely niche and just a "helper" for humans. They lack any creativity and an "update" requires a ton of resources. It's still fun to see how is it developing, it is impressive but kinda dubious at best.
For the free version, it won't let you delete your images that's a real bummer (unheard of)... especially since you can't keep it private but public by default.
3:00 number 2 just looks like a dating sim here lol "the lights look so beautiful today, protagonist-kun" pick a choice: A: not as beautiful as you. B: maybe we can go out at night more often then. C: yeah, looks pretty nice. D: *throw brick at her*
First place is both ideograph and flux pro, ideograph is more realistic but flux is more beautiful and creative. Mystic is second, and mid journey is clearly last
Unfortunately the takeaway is that from a commercial POV, as a graphic designer, I need to have at least 3 subscriptions. I've been with MJ for a while now, because up until recently there simply was no competition, particularly when it came to photographic quality. But I admit I have been very disappointed with the latest v6.1. I find it very hit and miss, the coherency is going backwards and so is the usual ‘hands & feet’ issue. Plus the censorship is getting out of hand, just because you want a “bare back” doesn't mean you are trying to create porn!!
Perhaps your text parsing wasn't clear. Perhaps you should have tried "warrior one" in quotes as to make it clear that "warrior one" is ONE ADJECTIVE and not TWO different words? Also, asking an AI program for something so specific that no one has probably entered it into the training data is not very fair. Warrior 1 pose? cmon man. Who is going to ask for that from an AI image program? This is like believing you are asking Jesus for a particular image and hoping to get back exactly what's in your mind's eye.
You probably don't use Midjourney but you should check the ? mark help thingy in the interface not just make things up mate.. relaxed, fast and turbo have no meaning for quality - you get unlimited relaxed hours where you lose priority but the fast hours (top priority) are capped. Turbo is faster yet but take more of your fast hours.
So I'm looking at this quality I use mid journey for hours a day I don't have problems with hands because you know why I use all the stylized references also you can bop down to version 6 use --s and up it to 500 or did you use --style raw for the photo's ?
while I like the idea of more competition in the field because it'll drive innovation, I'm also concerned about the possible practical ramifications. when there was "just" Stable Diffusion, everyone in the AI image community was working with these models and developing prompting styles / tools / loras / custom checkpoints on them. My concern is that with this diversification of underlying models, AI image generation will become a field entirely made up of niches, where it's increasingly difficult to find content for individual workflows. kind of like what happened to Javascript with the 10 trillion different frameworks that popped up.
I couldn't find the pictures on Twitter. It would've been nice to put the link to the direct comparison, instead of the generic link to your account. 👎
It’s funny you call her average. She looks 100 times better than the other generated photos. The other ones look far more static and styled than this one does yet, they made the woman’s face ultra cute and it feels like she’s one of those types of online influencer models.
24:57 these generations are purposely put out into the public to prevent a riot from either side after the end of this season‘s presidential election. In reality, the US greatness and stability comes from the fact that it can have totally different administrations empower one after the other, and the general characteristic and strategy of the country remains unchanged. In a strange metaphor, this is likely how our country has always ran
Not really. Midjourney struggles to follow precisely complex prompts 9 times out of 10. In any case, if you need to be specifically a Midjourney prompting expert, in order to get decent results, then that is already a fail when compared to other image generators.
Seems to me that Flux is still the best, out of the 4 models in this video, the google one I don't know much about after the original launch disaster, and dalle-3 I think might still be the top image generator, though Flux is very close.
Of course it does depend on how many rolls you do each time, what variables are used etc. as someone noted, one day comparisons will be irrelevant as all Ai generators will be near perfect.
I do find these A.I Channels funny. Constantly being amazed that the next a.i is better than the last. It's only ever going to get better than the last, kinda how A.I Works.
Well, we have done it. From now on I will not trust any picture any more. Yes, ok, if you know what to look for you can still spot little inconsistencies in AI made pictures, but they will vanish completely some day. Bah. I'd really preferred flying cars to this😊
You wrote hamala harris instead of kamala in prompt. That's probably why Mystic created the wrong image. Otherwise, thanks for the testing and effort... Ideogram and Mystic have greatly improved the eternal 'finger' problem.👍
I must correct you, that speed in Midjourney have nothing to do with quality. You can add --quality 2 or lower (default is 1) behind your prompt but the speed is only how much drain your GPU hours, if you use relax its unlimited but it takes longer and looks the same as if you use turbo or fast :)
![Eminem - Fuel (feat. JID) [Official Lyric Video]](http://i.ytimg.com/vi/t5H_CewqpKA/mqdefault.jpg)
🌏 Get an Exclusive NordVPN deal + 4 months extra here ➼ nordvpn.com/aisearch It’s risk-free with Nord’s 30-day money-back guarantee! ✌
👋
To🎉
Hey brother, what an amazing test you did!!!!!!! Can you tell us what is the best one for creating beards and hairstyles for men and good faces? Please!
If there's one thing I learned from this video, it's that Midjourney needs an update
Yea especially compared to imagen 3.
For realism yeah, but MidJourney blows all of them away if you want produce art, like paintings, abstract, creative. I think we need to forget about one tool for everything, I really hope MidJourney doesn't screw up it's style chasing photo realism.
@@TheFeedRocket Midjourney dormes not do the bestbart.. Niji does however.
If there's one thing I learned from this video, it's that Midjourney generates absolulely gorgeous women
@@TheFeedRocket yeah, you definitelly have a good point
I still prefer Flux. The Open Source fine tunes and tools, are still unmatched compared to the Closed Source models.
Still this is good. Competition breathes strength and like we've been saying at nauseam since SD1.5 burst into the scene.
This is the worst it'll ever get.
open source ftw!
Meh flux is actually kinda going openai way.
Schnell does not really produced serious production level albeit its fast and can be used for commercial.
Dev are not commercially available and pro well its just straight paying them the money.
Ideogram actually good but again it has too many limitation for the free tier.
Mystic is the same with that other existing one will run out of steam when we used it, unless you got a client who wants to create their own custom platform then yeah not worth it.
@@savire.ergheiz You can use images from Flux Dev commercially, just not the the model or Fine Tunes.
They put an enphase of distinguishing the images you generate, from the 'derivatives' (Read Fine Tunes, Loras and Merges). Likely not wanting to repeat the mess that was SD3 and it's Termes of Service.
@@vi6ddarkking Yea lets see later. We have so many examples of companies doing such thing like what Unity did. Its always beginning with simple limit like that.
True free open source won't really put any limits albeit it won't do the author any good rather than being famous.
@@savire.ergheiz It's not even clear that models themselves can be copywritten with the current rules. Much less the outputs. So the license is at best enforceable only on those using the model, and perhaps not even them. We'll have to see court cases on this to find out.
FYI - You misspelled "Kamala" as "Hamala", so it might have confused things for some of the generators.
ahhh shoot, good catch! I'll post some updated results on Twitter: x.com/aisearchio/status/1827420006612627545
In the image where it says "United We Stand" I laughed so hard 😂
BTW: for those who don't know, this guy is Facebook employee #1!!! =)
This is a great channel… but what we need is the leap from image generator to art directable images using Ai. That will be the true turning point… not what MJ does… but create a video of people walking down the street, then be able to select a tree in the image and say turn it 90 degrees, or make it back lit. Without that the generators are ‘you better like my first attempt, otherwise you’re stuck’. Vfx companies all know this… Sony is working with Sora trying to push this aspect. Hollywood is not going to get affected until this ability goes mainstream. I’m a vfx supervisor for movies, I love Ai… looking forward to the potential it offers but right now, it’s just good for ideas… after that it’s business as usual.
Thanks for sharing!
Use Florence-2, Flux, LLM and inpainting. Solved.
@@LArSON1942 not solved. Doesn’t work for video at the level needed.
"just good for ideas" what a simpleton.
Exactly 👍👍
Flux only costs your electricity though.
So true. haha😅
And you have to buy a super expensive GPU
That doesn't apply to the pro version (wich is significantly better than the ones you can run at home).
@@stephaneduhamel7706 go look at my civitai. you can add lora's to the dev version to make it much better
And is highly customizable!
One day these image comparisons will be pointless because they're all immaculate. Scary and exciting stuff
Midjourney FAST & RELAX do not affect quality; as FAST means you get the attention of the server almost immediately; and RELAX means that the server - when less busy - finally picks up your prompt!
Heavily disagree on ideogram being bad at anime. In fact, I like ideogram's Anime style more than the others.
Because it looks like actual anime scenes. With the others, I feel like, you can see that it's AI anime, which I don't like! I like the 2d look of ideogram's anime.
flux is still better because of open source and also we can use this in local pc and lora and other stuff and overall imagine 3 is best
open source ftw!
@@theAIsearch yeh
Good to see Midjourney being embarrassed with their low-quality crap. Hope this actually wakes them up and face the fact that no one cares about all the social stuff they're working on, instead of the actual model.
The prompts were low quality, so are the expected results. Midjourney outputs the same quality as the others you only have to know how to prompt is correctly.
VPN doesn't help with phishing, because the connection to the email server is encrypted by SSL. If your VPN is able to read your emails and "protect" you from phishing, it means that your data was somehow decrypted by vpn software and compromised. So, your advertisment is nonsense or it was anti-advertisement.
Yeah, strange since NordVPN's own website says that it doesn't protect from phishing.
VPN are pretty much useless those days because everything is encrypted at the application level.
The only use I see is to switch your IP to a different country to access servers that reject certain IP range, or serve different content.
@@xl000 VPNs are only becoming useless on the security side of things, they are still very relevant for firewall dodging, location spoofing, and anonymity.
We all knowns what VPN's are used for. Sailing the high seas.
It always was and as long as you can get cought doing just that, they will be useful. Sure, things like poor journalist is oppressive countries use them also. But I bet 98% of all VPN bandwidth is used for that type of acquired content.
@@ianPedlarTor browser is slow as fuck, so if you want to do stuff thay uses a lot of data like watch videos locked in your country, it's gonna a pain in the ass.
It doesn't matter how good the model is when it starts. What matters is whether you can train it and it can be improved by the community without censorship. So in this respect Flux has a bigger margin to develop itself. The marginal differences between these models is then not important anymore. Flux doesn't limit creation.
these AI image generators CLEARLY found something new recently for everybody to start pumping out stuff like this.
- 00:00 New image generators are released.
- 03:30 Image generation models test comparisons.
- 08:01 High-quality images need prompt specifics.
- 13:36 Mid Journey struggles with anatomy.
- 18:59 Ideogram excels at generating text.
- 26:01 Ideal results vary by model used.
- 34:28 Mid Journey fails in animal generation.
- 39:00 Anime styles differ between generators.
- 43:02 Context affects image generation accuracy.
- 45:16 Stay tuned for AI updates.
Summarized by GPT Breeze
I'm just waiting for a very stable one to help me with my creative work, and I can see, it's not too far, maybe by the end of this year.
This blind test is really good!
I love this style of video making.
Thank you!
Flux is superior just because the fact its open-source and you can bend it in any way and never says no. I wish more people would understand this. There is freedom without censorship. Be it politics, be it image generation or text generation.
2:52 [Image #1] - It's the Leoplurodon from that damn animated unicorn video... 'COME WITH US TO CANDY MOUNTAIN CHARLIEEEE!"
At this point, "better" doesn't mean much in terms of quality and adherence. Sure, there are some that are slightly better than others. But what it ultimately comes down to is censorship. I don't need a company telling me what I should be able to create. If do something "bad" with my creation, then I should be punished by the law of the land.
You're not wrong but that's not reality. These things are developed by people, and those people make the rules of what they develop. Not you. So not allowing them to censor their own product is taking away their freedom as well. No one is forcing you to use their product and play by their rules, but you want them to play by your rules.
@@drowzy2309 "So not allowing them to censor their own product is taking away their freedom as well" to me, that makes no sense. Why don't makers of kitchen knives make them all without a sharp point and a bit dull so that they can't be used to stab someone? Answer that and you should be able to answer the AI censorship issue as well. Childproof caps is another one. Good intentions, bad result. Now, for the sake of .1% of irresponsible parents and kneejerk idiots in power in govt, ALL parents, ALL people have to put up with struggling to open a pill bottle. At least now they have invented a DUAL top so you can flip it over and use it as a NORMAL top for those who do not have children. Point is, like Smokey The Bear said way back in the last century, matches don't start forest fires - people do. The tool should not be censored. The people should be punished only AFTER a crime has been committed as defined by each local govt. Companies shouldn't have a say in what other people MIGHT do with an object they made. Once it's created and released into the world, what Joe Blow does with it does NOT make the creators of it culpable in any way just like a match manufacturer isn't responsible for the arson. get it?
I feel like a big advantage Flux has is if there's something that it sucks at (like yoga poses or uncommon animals) you can just run it locally and train a LORA to make it better
I love FLUX for realistic stuff but fantasy or sci-fi... really hard to get
41:13 you completly missed the fact that the background in flux pro looks like real life instead of anime.
😂😂 The snail with legs from Midjourney was hilarious.
For the warrior yoga pose results : there is multiple warrior poses, the one you wanted is known as "warrior 1", while some of the results you got actually matched with other warrior poses, so i wouldn't exactly say that the ai is wrong.
hehehehe how come Pony is never mentioned around Anime... I know its very NSFW but its also very powerful in accuracy and the huge pervy community behind it that are really motivated to also make SFW Lora's
Midjourney only just having a web based interface is the most staggering news in this video.
Man those AI selfie girls are bangers lol
but as we all know, only open models with controlnets and lora features can be used in a professional way
open source ftw!
How is that? I don't see any reason why closed source models cannot be used in a professional way.
@@skyfe5430 you can but only for a few steps of an entire professionnal worfklow. Clients want specifics images with specifics composition and consistency, it's only achievable using controlnets. Midjourney/ideogram are good for creating nice images on instagram / creating ai video stories, book cover, but for the rest it's not that usable
If it's not open source I don't care
Same.
I use flux on discord I created a bot for it with shapes inc.
Yawn
open source ftw
11:00 and look at her right thumb. holy cow.
great comparison, but what you meant was manga, not anime (which are movies) and ideogram made actually the best manga art, also the letters on signs were kind of correct, it just was not english )) it was japanese alphabet katakana / hiragana
Hands and feet are usually the most challenging part for most models.
The title suggests that the comparison will be made to the maximum capacity of each tool. But this is not the case. When I see the importance of the "Guidance Scale" parameter on Flux Pro, I think you could have had much better results with a value of 2.5 (instead of 3.5). And this is certainly true for the other tools. Unfortunately (or fortunately) to know what a model has in its belly (French expression), you have to invest many hours. I'm still convinced that Flux is the most versatile and has the greatest potential for improvement, but... I'm going to take a look at Ideogram V2. Thank you so much fot this amazing work.
Thanks for sharing!
12:30 screw gewgull. they are already too huge and will soon take over the entire world.
First things first - on several occasions you said the hands on mystic looked good, when the were not. I ran the same prompts on Flux-dev, running locally and I have to admit, I was surprised that it didn't do the hands/feet prompt that well. Also, when removing the "dragon" word from the komodo prompt it did comparably well as midjourny in your case. just the tongue wasn't quite right. All other prompts where either absolutely perfectly following the prompts, or where very close to perfect. I'll definitely try ideogram, as it really stands out in your tests.
Informative video, but the fact is that there are different ways to prompt for different models, and also, there are different tools on each website. For example, I would never even try to prompt for text on Midjourney.
Also in Midjourney, you can use --niji for anime-specific model. And for low quality, phone selfie image, try using --s 0
The new Imagen 3 actually does by far the most realistic human anatomy features than any other model out there. It's a lot better than Ideogram 2, Mystic, Flux, etc. Imagen 3 finally got hands and feet right. Literally the human anatomy on Imagen 3 is beyond any of these and nobody is talking about it. That is quite ironic and quite hilarious actually.
It's censored af
yeah, Imagen 3 is underrated & free!
“Imagen 3 is by far the best at human anatomy.”
Imagen 3: _Can’t even generate a boob._
@@nemonomen3340 I'm a simple man. I try out various AI image generators and image to video generators and my number one question to all of them in my mind is, "Are you going to let me generate some huge boobs or not?"
@@nemonomen3340 well I don't disagree with that part lol it's the most censored model too, hence why nobody cares
I think the tests should be small variations of previous tests, generators could easily incorporate the previous failed prompts in their fine tuning, so cat could be on top if a yellow cube with a red square behind, every time a small variation on the same theme..
Midjourney is the only one that has "character" and "style" references. A great image I can never imagine again makes it good for single use cases only.
Just hoping to run it locally in the future. good video
Thanks for comprehensive run down. Subbed)
Thanks for the sub!
you should've included Leonardoai, In my opinion, it's awsome too.
A couple of good tests for you: Man eating a banana - MJ doesn't peel it, MS does. Waterskiing - both have ropes all over the place. Which is worse though? Probably MJ.
Number one mid journey has had a website it just hasn't been open for everyone and number two. The speed on mid journey determines how fast you are in the queue to be rendered it does not affect the quality at all. It will actually take the same time to render for all choices.
I think open source is best cause you can train them as you want, You get so much functionality to develop those models. The way you want, You get a control net, an ip adapter, inpaint option , and If you know coding you can make those models the work way you want and can sell those specially built model
i noticed the pattern for midjourney is they use ban on users to basically deny service while keeping your money that you paid to them.
its a very strange pattern that i uncovered when discussing with people who paid for annual subscriptions and got banned because of like what you encountered, triggering the NSFW over images that shouldn't really be an issue.
thanks for sharing
If you test Midjourney anime you should be using Niji 6, not MJ6.1. Niji 6 is a really good anime generator (though it has much less knowledge of IPs than some like NAI). Also, you can get around MJ's moderation by using words like 'president' - because it strongly believes we have only ever had one president, and it was the 45th - but you'll likely end up with him kissing Hillary, rather than Kamala.
try number 2 at asking AI Search to prompt "4 capybaras stacked on top of one another, realistic"
for the selfie photo perhaps we should mention which generation phone the selfie was taken 😋 .. the modern phone selfie portrait mode is really good so no surprise flux-pro generated it with all the background blur 😀
One of the better videos on the topic, for sure. Most are biased and/or obnoxious presenters. All in all, I just get more confused every day as the list of image makers grows and each one claims to be The New Best One Ever. The text (prompts) remind me of old text-adventure games-you knew what action you wanted, but the text parser was so limited that it took ages, blindly guessing what verbs and nouns the game had in its vocabulary-any others just got an "I don't understand" response. Today, between text and imagery is a real disconnect between how much "AI" in general can comprehend, much less synthesize.
Thanks for sharing!
Ideogram is definitely the best one by far
Always really appreciate your videos, thank you! Ideogram is exceptionally good at working with text, really represents a big step forwards.
Here commenting to help you out! Lets go! You got this! 🎉
Thank you!
My choices: Sometimes I choose Flux Pro, sometimes Midjourney... None for Mystic and only two times Ideogram.
Thanks for sharing!
To this day im so shocked why freepik doesnt get more shoutouts.
100 bucks a year.
200000 credits.
Flux, mystic, magnific upscaler. Sketch to AI. RE IMAGINE, INPAINT, EXPAND...
PLUS 1 million stock images, templates
Plus video stock.
Its a bohemoth !!!
I pick 2 looking the best, 4 was pretty good too, some more creative prompts i liked 3. 2 ended up being Ideogram v2, 3 is midjourney
Of all your prompts in the blind test I ended up choosing #4 most of the time. Turns out that was FLUX.
24:00 Kamala, not Hamala 😅. You can see Magic Prompt contains corrected spelling.
It's CumAllah
33:15 the dog looks super sharp but has two front paws 🐾 on the floor and a third on the cube. 😅
The Flux version for the verify me only flaw was how good it was, and I wonder if you took the smaller versions and did the same thing could you tweak the low quality out a bit more. But what really caught my idea on it was how well the sign was done in low quality. The lettering isn't perfect like a font, the f is redrawn over like it was really hand made. Take that photo and run it through a sharpen filter and I think you'd have 100%.
Every week in AI in advancements is like a year in previous tech advancement duration
I've been enthusiastic about AI for a while I think is a good thing to have but I can't help but worry as the "why" of all of this?
Specially considering how much money is being put onto this. As of now AI models seem to be only good at an extremely specific one thing and in order to reach it they have insane requirements.
Even if they come out as a fully realized working machinery I can't help but think that they will be extremely niche and just a "helper" for humans.
They lack any creativity and an "update" requires a ton of resources.
It's still fun to see how is it developing, it is impressive but kinda dubious at best.
39:50 Just personally, she reminds me of Mirai from Kyoukai no Kanata.
For the free version, it won't let you delete your images that's a real bummer (unheard of)... especially since you can't keep it private but public by default.
To really compare the power of the models you should set them to quality or best mode if available.
3:00 number 2 just looks like a dating sim here lol
"the lights look so beautiful today, protagonist-kun"
pick a choice: A: not as beautiful as you. B: maybe we can go out at night more often then. C: yeah, looks pretty nice. D: *throw brick at her*
number 2 just looks like a dating sim here lol
This is clearly a trick choice and she has a pistol locked and loaded in her pocket, ready to mug you the moment you look away so it has to be D.
First place is both ideograph and flux pro, ideograph is more realistic but flux is more beautiful and creative. Mystic is second, and mid journey is clearly last
Unfortunately the takeaway is that from a commercial POV, as a graphic designer, I need to have at least 3 subscriptions. I've been with MJ for a while now, because up until recently there simply was no competition, particularly when it came to photographic quality. But I admit I have been very disappointed with the latest v6.1. I find it very hit and miss, the coherency is going backwards and so is the usual ‘hands & feet’ issue. Plus the censorship is getting out of hand, just because you want a “bare back” doesn't mean you are trying to create porn!!
Perhaps your text parsing wasn't clear. Perhaps you should have tried "warrior one" in quotes as to make it clear that "warrior one" is ONE ADJECTIVE and not TWO different words? Also, asking an AI program for something so specific that no one has probably entered it into the training data is not very fair. Warrior 1 pose? cmon man. Who is going to ask for that from an AI image program? This is like believing you are asking Jesus for a particular image and hoping to get back exactly what's in your mind's eye.
😁😆 that badfinger at 15:20 🤣🤣🤣
what the hell happened to midjourney
Flux is still the best overall.
You probably don't use Midjourney but you should check the ? mark help thingy in the interface not just make things up mate.. relaxed, fast and turbo have no meaning for quality - you get unlimited relaxed hours where you lose priority but the fast hours (top priority) are capped. Turbo is faster yet but take more of your fast hours.
Well, as someone who has used many AI generators, still Midjourney is top-notch, although they do have shortcomings...
So I'm looking at this quality I use mid journey for hours a day I don't have problems with hands because you know why I use all the stylized references also you can bop down to version 6 use --s and up it to 500 or did you use --style raw for the photo's ?
while I like the idea of more competition in the field because it'll drive innovation, I'm also concerned about the possible practical ramifications. when there was "just" Stable Diffusion, everyone in the AI image community was working with these models and developing prompting styles / tools / loras / custom checkpoints on them. My concern is that with this diversification of underlying models, AI image generation will become a field entirely made up of niches, where it's increasingly difficult to find content for individual workflows. kind of like what happened to Javascript with the 10 trillion different frameworks that popped up.
I cannot find the Glif / Build link...
does this work glif.app/@Me3/glifs/clzuhyld9000a81hplo04cy1h/source
"I'm not gonna tell you which image is generated by which model"
You always put the newest model on the left
Yeah but please don't forget King for image generation. I think it's awesome and I'd like to see it compared and what you think 😊
Model 4 looks amazing. Which one is that?
not sure I'd say Mystic is better than Flux, but Ideogram certainly is. very impressive.
it was later revealed that mystic was just an upscaled wrapper of flux, which was disappointing
39:52 She looks like that girl from a silent voice
aha! that's the one
for the blind test my favorite changed pretty much from prompt to prompt. except 1, 1 just sucks
I feel like we need an update on more Free image gens that are similar to seaart, yodayo etc. in whcih u cam choose lora and model
I couldn't find the pictures on Twitter. It would've been nice to put the link to the direct comparison, instead of the generic link to your account. 👎
Mystic and Flux are definitely using the same training data IMO. They all default to the same tendencies in multiple secenarios
It’s funny you call her average. She looks 100 times better than the other generated photos. The other ones look far more static and styled than this one does yet, they made the woman’s face ultra cute and it feels like she’s one of those types of online influencer models.
And I’m talking about ideogram.
24:57 these generations are purposely put out into the public to prevent a riot from either side after the end of this season‘s presidential election.
In reality, the US greatness and stability comes from the fact that it can have totally different administrations empower one after the other, and the general characteristic and strategy of the country remains unchanged.
In a strange metaphor, this is likely how our country has always ran
35:33 you trashing mid journey is giving me life 😂
Time to remove my subscription come to think of 🤔
To be fair Midjourney could have created the same quality as the others. It is all about the prompts.
Not really. Midjourney struggles to follow precisely complex prompts 9 times out of 10. In any case, if you need to be specifically a Midjourney prompting expert, in order to get decent results, then that is already a fail when compared to other image generators.
32:28 this is a cat!
flux rocks, it came out of nowhere!
we have really good chat ai's, really good music ai's, really good video ai's, and very good image ai's, what's next?
Seems to me that Flux is still the best, out of the 4 models in this video, the google one I don't know much about after the original launch disaster, and dalle-3 I think might still be the top image generator, though Flux is very close.
Of course it does depend on how many rolls you do each time, what variables are used etc. as someone noted, one day comparisons will be irrelevant as all Ai generators will be near perfect.
None of these things can do legs. Ask for an image of someone wearing black shoes and you get nothing below the waist.
I do find these A.I Channels funny. Constantly being amazed that the next a.i is better than the last. It's only ever going to get better than the last, kinda how A.I Works.
its neverending!
Well, we have done it. From now on I will not trust any picture any more. Yes, ok, if you know what to look for you can still spot little inconsistencies in AI made pictures, but they will vanish completely some day.
Bah.
I'd really preferred flying cars to this😊
You wrote hamala harris instead of kamala in prompt. That's probably why Mystic created the wrong image. Otherwise, thanks for the testing and effort... Ideogram and Mystic have greatly improved the eternal 'finger' problem.👍
good spot. i tested them again with 'kamala' but interestingly, they still couldn't generate her: x.com/aisearchio/status/1827420006612627545
I must correct you, that speed in Midjourney have nothing to do with quality. You can add --quality 2 or lower (default is 1) behind your prompt but the speed is only how much drain your GPU hours, if you use relax its unlimited but it takes longer and looks the same as if you use turbo or fast :)
Which one is good for making assets for web design?
Midjourney is not meant for anime, technically their Niji version is meant for it.