Again a very good video, Matt! There is however one remark I'd like to make. Through working with some of these tools myself, I have learned that prompting for different models requires some differences in the prompt itself. So ... once you choose for one or more tools, you need to learn it's characteristics and prompt accordingly. (The biggest difference is in using e.g. Midjourney and Stable Diffusion) The same is true with LLMs by the way. And quite honestly, the same is true for talking to people ;-)
You should incorporate a censorship category into your analysis. I tried that photo shop AI and it censored nearly everything i threw at it so i bailed. It wouldn’t even allow me to render a soldier holding a rifle.
@@thepde6957 This is exactly why I’m considering trying Grok. I don’t care if it’s the best or not. I just want something that doesn’t censor me constantly for completely innocent things and threaten to suspend my account because I’m tying to make an image of a soldier holding a rifle.
I really want to see more testing of non-realistic outputs. Things like illustrations, vector art, comics, watercolour, painting, visual illusions, etc
MJ is much deeper in terms of aesthetic exploration, but they all do reasonable photo realism now . I still feel MJ has more creative originality with style references and personalisation codes but it’s not the most accessible and it’s pretty expensive. Horses for courses I guess.
@natevelar In my experience, midjourney has really good inpainting features. You can consistently inpaint to fix things up. Unlike some others I've tried, trying to fix small stuff turns into a nightmare because rather than fixing something, it ends up turning into something unrelated to the overall image. Also, the style and character reference was one of the big things I liked about it. It is very easy to use. I've tried Grok 2 recently, but it didn't look all that good tbh. No inpainting feature as well. Like Dall-E but with less features. Krea ai also has flux and looks a lot better while also being fully uncensored. I haven't tried original flux yet. Too many out there and no time to try them all.
you should calculate quantitatively how adherent each is to the prompt. I have done my own and have run each model four times with the same prompt, then tallied the number of times each model captured the requested elements. For relatively simple realistic prompts I got Imagen 3 with 93% adherence, tied with ideogram ((93%) followed by Auraflow (88%), Leonardo Phoenix (85%), Flux Pro (81%) ´DALL-E 3 (78%), Meta AI (74%), Amazon Titan v2 (64%)
Why be blackmailed by the retro Mid J; I get over 200 images totally free virtually every day from Fotor, Leonardo, Dzine, Sea Art, Night Cafe, OpenArt, Tensor....and more. And all without the discord chit.
Midjourney really needs to boost generations for paying customers. I was trying to render a bunch of images to create some side views of a character I want to be consistent and ran out of generation amounts ... I paid a bit more for more generations and ran out again pretty quick.
It seems like the best way to ensure your model remains competitive is to get rid of the content filters. Since so many of these models are heavily filtered, you could definitely ensure market share by removing them and letting people create whatever they want. The content filters are a little silly, in my opinion. This applies to all AI models, especially ChatGPT, which is notorious for being absurdly strict to the point of blocking legitimate requests that are not harmful in any way.
It's a good business strategy until you get sued into oblivion because your unrestricted model allowed some high schooler to set off a pipe bomb at assembly
@@wishesandfishes Does google get sued for showing fake images? no. Does a gun manufacturer get sued when someone uses the gun to kill someone? no. If you use photoshop to smear a political candidate, is photoshop responsible? no. When it comes to individual liberty, there must be individual responsibility. Don't blame the tool, blame the user. Regardless, AI is so much better at recognizing intent now that you could easily block obvious cases of misuse while letting a lot more through. The idea that ERP is still not allowed on ChatGPT and other models is honestly ridiculous. I get an age filter, but this is something that users should be able to opt in and consent to and use freely. Same goes for many other things.
MJ's quality exceeded others for so long but their implementation of use was awful, the whole discord thing hurt it more than helped it and they were just slow creating updates whilst complicating prompting methods. I Bounced in V3.
@Bhodisatvas but they have a whole working website since around V5 now(was for members with high generation counts for a little while), and they're at V6.1 currently, which is accessible to pretty much anyone. If you bounced since V3, which was way in the early days, you can't really complain. AI image generation was still new during that time, and no one could generate an image of a person without it looking like a kid's drawing from any AI at that time, lmao.
@@barronhelmutschnitzelnazi2188 I've been following MJ ever since and nothing has tempted me to return. The prompting method turned into it's own language and I'd rather use other generators that can understand sentences without having to add a ton of symbols and numbers. MJ fell behind in usability and cost effectiveness. It is a shame because I absolutely adored the feel of the images and as an artist I get it but their competitors have raised the bar.
I paid for pro through this year and I kind of regret it. They never gave us an API, photo styles all look undercooked, it fails at fine details. And you still cannot delete images using a straightforward interface. It’s a toy in its current state, good for finding a look but that’s about it. Open source is king.
Hey Matt, a great comparison you did here! Something that really gives a map of current image generation landscape! As for additional tests, would be great to test different styles as it’s where the leaders seem to stand out, and also a ‘gimme five’ test for hands consistently 😅
Matt, try these categories: - a military vehicle or mech, especially large scale. - a building or compound with a top down view. - cartoon character with accurate hands + painted fingernails. Usually image generators struggle a lot with those.
I love the idea of a figma board in real time that discuss prices, sensorship, abilities and others like that. Sometimes Its starting to be hard to keep up w all models available and manually check everyone seems like a hazle. Thanks for the video Matt!
I found with Midjourney to create a surreal image like your octopus image you need to adjust the --weird setting to get it to do it, otherwise it tries to create a realistic image.
Please add some comic scene, game art or anime prompt to this comparison. I really like these comparison videos. This helps me decide where I should invest my money when developing my game. Thanks a lot
Miggy Awistss sounds even cooler than just "Magic Awaits" because it's like either something is trying to tell you Magic Awaits, but it doesn't get English quite right OR you can't really read the magic words very well.
Thanks for posting. Between learning ComfyUI/Flux and Grok, forgot about Leonardo. MJ still has that finished quality on images, if not text. No longer bothering with Firefly.
I had the exact opposite opinion on the face realism. The Firefly image is the only one that looked photo realistic to me. The rest of them have far too much contrast in the face, they look distinctly AI generated.
I still feel that MJ has the best artsy feel. I am not much interested in photorealism, which I found relatively boring to be honest. And in that I haven't yet anything even remotely as good as MJ by far.
Same. I still think MJ is vastly superior to others simply because it allows getting images in very specific styles. The fact you can show MJ an image of a particular style and is able to replicate it is truly incredible. Realism is vastly overrated anyway.
If you haven't seen anything as good as MJ yet, you must be using the most barebones prompts ever. Midjourney can't seem to get anything right when I use it. It has been a while, though.
Excellent Figma Board comparisons, thanks Matt! Can you do prompt comparisons that show off the ability to do non-photorealistic styles? Like watercolor, sketches, thick brush strokes, abstract art, impressionism, etc.?
One of the biggest public use cases for image generation is fantasy images for characters in D&D campaigns, or art to accompany other creative works like writing etc. I think Midjourney probably still wins, but that is one thing it is worth testing when you compare. It is still pretty difficult to give a detailed description of a unique character -- like a fantasy hero or villain that you invent yourself -- and get results that actually match your vision for it.
The best is Leonardo! the one I use for almost everything, but for some pictures that this and others AI censure, I really prefer to use Seaart AI, I wonder why is not in this video. It is really amazing and one of the few ones who has incorporated Flux 1.0 pro already
I am very, very new to AI image generation, and this video helped quite a bit! I still not sure what to use, as I have access to firefly through my adobe sub. But I'm looking into investing into something. Thank you for this video!
Thank you for this roundup, the side by side visual was a brilliant way to really see how they compare. I already subscribe to Leonardo (have to support my fellow Sydney siders) and might have to pay for ideogram soon too as it looks fantastic and gave me some excellent new variations on a project I am working on.
Midjourney always seems to prioritize generating cinematic, aesthetically pleasing images over prompt adherence. Which might be an advantage in some situations
Firefly credits are included in your Adobe account. The amount of credits depends on the subscription to Adobe you have. Full suite 1000 credits or just Photoshop 250. These are also the credits used for gen fill. You get 1500 credits if you also have Adobe Stock, but that's dumb because of the cost and you can make the images instead.
I play around with a few webpage based models but at the end of the day I just boot up a local version of Stable Diffusion. It does everything I need it to do for images.
Matt, I can’t get Leonardo to create an old man looking at his younger self in a full length mirror. Ideally holding an electric guitar. I just get totally random responses.
Thanks Boss BG ❤ so much work shared with us to help us and grow our knowledges 😁 yk I talk about your channel always 💯 for me , si far away, but weirdly so near, you are just the best 🔥🔥🔥
It’s funny when someone like Ideogram tries to work the word Idea into their name, but then you pronounce it like they wanted the word “Idiot” in their name 😂 Another excellent video about all the fun tools we have to play with 🎉
Midjourney needs to lower costs substantially, or they are going to be done for. Or at least give us a huge update that beats Magnific or something to justify it now.
😆 🤣 😂 no they're not........ but the stiff competition will hopefully make the devs stop screwing around with wasteful tasks such as developing unnecessary extra robust community interaction tools, etc, etc, etc. If I want to interact with community...... that's what Discord and social media is for. I just want the best AI image tool....... period.
Now that we have all of these free image generators, how long will it be before someone combines the best examples into a meta generation tool, which allows you to select the best images from five different models?
Wow that Black Forest Ham reference when talking about going Ham on Flux which is made by Black Forest Labs 💀😂🤣 I think it went over everyones head 😅 0:26
I found flux schnell the other day on Poe. Was so impressed I unsubscribed from MJ. I just could not ever get my words right to create what was on my mind in Midjourney. Flux understood me very well and it didn’t take hours of tweaking to get what I needed. ❤️🇦🇺
It doesn't really follow the prompt though - where are the deep wrinkles? He doesn't seem particularly weathered, and the sea isn't that stormy either.
it would be neat if on the Figma board, you had all the test images for each model, so for example the 4 images for the old man sea photo for Ideogram arranged in a 2x2 grid, then users could zoom in to see the full quality of each I've never cared too much about perfect AI text rendering cuz you get much finer control using photoshop, as long as there's a space for the text and a background that isn't too noisy, it's fine
Ideogram is way better at prompt adherence than Dall-E3. Most text to image generators blow up if you ask for two distinct characters in the same prompt, and will often try to use them together (for example if you put Red Riding Hood and the Big Bad Wolf, they will give Red Riding hood fur). But Ideogram does not have this issue, Flux is terrible at it, and Dall-E3 can do it but the context window prevents you from having specifics.
The newer models seem to be limited to more generic styles, and trying a combination of named artists often is ignored. That’s one reason I’m still with MJ.
1.) Ideogram is the new king of the hill! 2.) Mystic has Flux underneath. 4.) Ideogram + Kling + Udio + Capcut 😂 = your personal Hollywood production company.
Very cool tutorial. Thank you so much for this. I'm a huge fan. I had a question about commercial rights. Do all them allow use commercially? I mainly use AI for print on demand so I would love to see examples or prompts for a flower pattern that's tiled, a cool sticker, a coloring book style design where it's black and white and a t-shirt design with text and a graphic with a black background so that I can remove it easily. Oh... I would also like to know the size. Are they all 1024x1024. Can I upscale them in app?
You should incorporate a infografic category, most of the models can't put details, text or another smart little elements or texts in to one single image, Ideogram 2.0 is the most impresive I have seen in doing that, and any other model can't even get close. Great to make a ppt
Midjourney is still quite a long way ahead when prompting “in the style of”. The different styles and their aesthetic appeal is way beyond any other model including Flux. That said I think their business model will need to change. Ideogram, Dalle and free huggingface Flux generations can do most things most people would need for free.
It kind of seems like Midjourney Adobe and even runway prices are a little high. Hopefully they will make it a little more reasonable to compete with others. Competition can be good. I don’t think Midjourney free trial will help
Leonardo ai is great and I prefer it over ideogram. While ideogram does the best with spelling of the words, but I think Leonardo produces much better images over all. As for Midjourney typography, it's disappointing that almost everything produced is misspelled and very small. A lot of the times on Midjourney the images show up behind characters or objects and this needs to be updated asap. Also I get frustrated with Midjourney's prompting, it's time they start making massive improvements. I like all three ai image generators and recommend each of them. Also they all have improved on the problems of hallucinations, specially on the hands.
I will never use Midjourney. They showed their true colors by being greedy while others were opensource and free. Now they want to backtrack and make it seem like they are doing a good thing, but they were just hindering artists trying to get their foot in the door with their paid options. Way better options now and I hope they get left behind in the dust.
You can generate all kinds of images on all kinds of platforms, ultimately the difference will be which users are profitable to these companies long-term and who will find uses for the work they create. It’s an expensive and very time-consuming hobby for the majority of users right now. How long people pursue something that requires so much personal resource without obvious financial return might ultimately be where AI image generators hit the buffers. Anything that becomes free and everyone can do will ultimately become devalued once the novelty goes.
I guess what the user sees in the image matters a lot too. To me, the only one from the realistic/people category that actually looks like a real image is... Firefly. All the other, with the possible exception of Midjourney, are so painfully obviously fake that it hurts to look 😬 And I don't mean the images are bad, they're not, just that they scream AI generation
In addition, I have never got a good image from Leonardo's own models to the point that I've stopped even trying them out completely. Way, way too much contrast in every image, to me it spoils them completely. I'll reiterate that I don't mean the images are somehow universally bad, it's just so heavily dependent on the user. It's a blessing that we have many models to choose from, so everyone can decide for themselves!
AI Image generators are getting better. I will be coming back to this video, but I know it will probably be obselete in a year or maybe even a few months. Then it will be time for you to do a new version of this video which is great. lol
I have to disagree with Firefly. I think they have the most realistic image of the old man and sea (although you can't really see the sea). I wouldn't think its AI made ...
I think mid journey is really good at art styles, I’m also a pay user of Leonardo but it has difficulty of generate image with it’s style reference. Besides, I trained a model in Leonardo with my own photos and it just generate my “photo” whatever I try the prompts. Suggestions?
I signed in with google and it still says I need a subscription to do anything or search for anything, didn't see anything about a free trial. Lots of vids and articles saying it's free but none with any useful information on how to do that, including this one.
MJ is bad at prompt adherence but it produce the image so beautiful that made the user forget what their prompt was. The advantage of MJ is don't care about the prompt as long it produce the ultra beautiful image.
Again a very good video, Matt!
There is however one remark I'd like to make. Through working with some of these tools myself, I have learned that prompting for different models requires some differences in the prompt itself. So ... once you choose for one or more tools, you need to learn it's characteristics and prompt accordingly. (The biggest difference is in using e.g. Midjourney and Stable Diffusion) The same is true with LLMs by the way. And quite honestly, the same is true for talking to people ;-)
You should incorporate a censorship category into your analysis. I tried that photo shop AI and it censored nearly everything i threw at it so i bailed. It wouldn’t even allow me to render a soldier holding a rifle.
Same here. I'm working on a military boardgame counters and everything is censored,
@@thepde6957 This is exactly why I’m considering trying Grok. I don’t care if it’s the best or not. I just want something that doesn’t censor me constantly for completely innocent things and threaten to suspend my account because I’m tying to make an image of a soldier holding a rifle.
Photoshop is pretty much the worst at that sort of thing. It hits false positives pretty frequently as well
Excellent suggestion
Right too much censoring sucks ,
I really want to see more testing of non-realistic outputs. Things like illustrations, vector art, comics, watercolour, painting, visual illusions, etc
Why not testing it for yourself?
Agreed that would be cool😊
Yep , very good Idea but he can't do all our use cases 😉 perhaps someone of our comu could propose a work, plz just do it 😊 🕊️
Agree. Image generators shouldn't be limited to images that look either hyper-realistic or look like stock illustrations.
I hope they dont do art well
MJ is much deeper in terms of aesthetic exploration, but they all do reasonable photo realism now . I still feel MJ has more creative originality with style references and personalisation codes but it’s not the most accessible and it’s pretty expensive. Horses for courses I guess.
Competition is good! Keep Midjourney on their toes. Mystic is KILLING IT!
which one do you think it best at the moment?
@natevelar In my experience, midjourney has really good inpainting features. You can consistently inpaint to fix things up. Unlike some others I've tried, trying to fix small stuff turns into a nightmare because rather than fixing something, it ends up turning into something unrelated to the overall image. Also, the style and character reference was one of the big things I liked about it. It is very easy to use. I've tried Grok 2 recently, but it didn't look all that good tbh. No inpainting feature as well. Like Dall-E but with less features. Krea ai also has flux and looks a lot better while also being fully uncensored. I haven't tried original flux yet. Too many out there and no time to try them all.
you should calculate quantitatively how adherent each is to the prompt. I have done my own and have run each model four times with the same prompt, then tallied the number of times each model captured the requested elements. For relatively simple realistic prompts I got Imagen 3 with 93% adherence, tied with ideogram ((93%) followed by Auraflow (88%), Leonardo Phoenix (85%), Flux Pro (81%) ´DALL-E 3 (78%), Meta AI (74%), Amazon Titan v2 (64%)
so... Midjourney is STILL NOT FREE. Got it.
Or no ... After four generations my daily credits were used up.
It shouldn’t have taken this long for Midjourney to release their website. They really have serious competition now and I wonder if they can handle.
Why be blackmailed by the retro Mid J; I get over 200 images totally free virtually every day from Fotor, Leonardo, Dzine, Sea Art, Night Cafe, OpenArt, Tensor....and more. And all without the discord chit.
📸 ty
Is there a meta tool that allows you to choose from multiple generations from different models yet?
All of them generate trash compared to MJ.
after MidJ which do you think is the best?
@@johnhawkinshawkins1284 Don't forget huggingface. Some unsensored ones on there too
Midjourney really needs to boost generations for paying customers. I was trying to render a bunch of images to create some side views of a character I want to be consistent and ran out of generation amounts ... I paid a bit more for more generations and ran out again pretty quick.
Got a runaway 3 feeling here?
It seems like the best way to ensure your model remains competitive is to get rid of the content filters. Since so many of these models are heavily filtered, you could definitely ensure market share by removing them and letting people create whatever they want. The content filters are a little silly, in my opinion. This applies to all AI models, especially ChatGPT, which is notorious for being absurdly strict to the point of blocking legitimate requests that are not harmful in any way.
If it comes up in a Google image search or is on wikipedia, it should be ok for ai. Just as a "safe" mode and age tiers.
it sounds good, but it's also Pandora's Box, whacked out teens giggling over fooling people with faked reality photos could become a hobby or sport
It's a good business strategy until you get sued into oblivion because your unrestricted model allowed some high schooler to set off a pipe bomb at assembly
@@wishesandfishes except why isn't Elon Musk concerned about lawsuits from realistic fake images from Grok 2.0
@@wishesandfishes Does google get sued for showing fake images? no. Does a gun manufacturer get sued when someone uses the gun to kill someone? no. If you use photoshop to smear a political candidate, is photoshop responsible? no.
When it comes to individual liberty, there must be individual responsibility. Don't blame the tool, blame the user.
Regardless, AI is so much better at recognizing intent now that you could easily block obvious cases of misuse while letting a lot more through. The idea that ERP is still not allowed on ChatGPT and other models is honestly ridiculous. I get an age filter, but this is something that users should be able to opt in and consent to and use freely. Same goes for many other things.
MJ's quality exceeded others for so long but their implementation of use was awful, the whole discord thing hurt it more than helped it and they were just slow creating updates whilst complicating prompting methods.
I Bounced in V3.
@Bhodisatvas but they have a whole working website since around V5 now(was for members with high generation counts for a little while), and they're at V6.1 currently, which is accessible to pretty much anyone. If you bounced since V3, which was way in the early days, you can't really complain. AI image generation was still new during that time, and no one could generate an image of a person without it looking like a kid's drawing from any AI at that time, lmao.
@@barronhelmutschnitzelnazi2188 I've been following MJ ever since and nothing has tempted me to return. The prompting method turned into it's own language and I'd rather use other generators that can understand sentences without having to add a ton of symbols and numbers. MJ fell behind in usability and cost effectiveness. It is a shame because I absolutely adored the feel of the images and as an artist I get it but their competitors have raised the bar.
I paid for pro through this year and I kind of regret it. They never gave us an API, photo styles all look undercooked, it fails at fine details. And you still cannot delete images using a straightforward interface. It’s a toy in its current state, good for finding a look but that’s about it. Open source is king.
@@renderartist It also still fails miserably at text. It is falling further and further behind.
Love your channel, Matt. some prices below each row would be cool. just a small suggestion.
Hey Matt, a great comparison you did here! Something that really gives a map of current image generation landscape! As for additional tests, would be great to test different styles as it’s where the leaders seem to stand out, and also a ‘gimme five’ test for hands consistently 😅
I’ve always used Ideogram especially for graphic design since you talked about it. This new model is simply a beast.
Matt, try these categories:
- a military vehicle or mech, especially large scale.
- a building or compound with a top down view.
- cartoon character with accurate hands + painted fingernails.
Usually image generators struggle a lot with those.
I love the idea of a figma board in real time that discuss prices, sensorship, abilities and others like that. Sometimes Its starting to be hard to keep up w all models available and manually check everyone seems like a hazle. Thanks for the video Matt!
This video is stupid useful. Thanks!
Also, the AI face that you use in your videos looks soooo good and realistic!
Even your fingers look realistic. What is the AI tool that you are using to generate the guy witb the mic/headphones?
I found with Midjourney to create a surreal image like your octopus image you need to adjust the --weird setting to get it to do it, otherwise it tries to create a realistic image.
Please add some comic scene, game art or anime prompt to this comparison. I really like these comparison videos. This helps me decide where I should invest my money when developing my game. Thanks a lot
Miggy Awistss sounds even cooler than just "Magic Awaits" because it's like either something is trying to tell you Magic Awaits, but it doesn't get English quite right OR you can't really read the magic words very well.
Yeah, there is something magical about that AI tardness...
Awesome comparison. I was totally confused as to what was the real purpose of all these image generators dropping every week . Thanks❤🔥
Thanks for posting. Between learning ComfyUI/Flux and Grok, forgot about Leonardo. MJ still has that finished quality on images, if not text. No longer bothering with Firefly.
I love when you test them against each other. I have been doing it myself finding the best picture that way! I love your channel!
I had the exact opposite opinion on the face realism. The Firefly image is the only one that looked photo realistic to me. The rest of them have far too much contrast in the face, they look distinctly AI generated.
I was thinking the same thing
I agree. The others are just too 'over processed' or dramatic. The Firefly face is close to being a regular person.
I still feel that MJ has the best artsy feel. I am not much interested in photorealism, which I found relatively boring to be honest. And in that I haven't yet anything even remotely as good as MJ by far.
Ideogram version 2, that just came out, does the best paintings I’ve ever seen, and it’s not even close
Yes especially on lighting quality.
Same. I still think MJ is vastly superior to others simply because it allows getting images in very specific styles. The fact you can show MJ an image of a particular style and is able to replicate it is truly incredible. Realism is vastly overrated anyway.
Img to img flows are crispy on MJ. Never had a good enough tool for consistent, customizeable, styles
If you haven't seen anything as good as MJ yet, you must be using the most barebones prompts ever. Midjourney can't seem to get anything right when I use it.
It has been a while, though.
Excellent Figma Board comparisons, thanks Matt!
Can you do prompt comparisons that show off the ability to do non-photorealistic styles?
Like watercolor, sketches, thick brush strokes, abstract art, impressionism, etc.?
The breakdown of AI image generators was perfect 👍 Thank you!
Your sigma board is in my favorites, I really want to be able to compare these models, especially in the photo realistic area.
One of the biggest public use cases for image generation is fantasy images for characters in D&D campaigns, or art to accompany other creative works like writing etc. I think Midjourney probably still wins, but that is one thing it is worth testing when you compare. It is still pretty difficult to give a detailed description of a unique character -- like a fantasy hero or villain that you invent yourself -- and get results that actually match your vision for it.
Thank you Matt! What a great summary of the current top AI image generation tools. I subbed!
The best is Leonardo! the one I use for almost everything, but for some pictures that this and others AI censure, I really prefer to use Seaart AI, I wonder why is not in this video. It is really amazing and one of the few ones who has incorporated Flux 1.0 pro already
I am very, very new to AI image generation, and this video helped quite a bit! I still not sure what to use, as I have access to firefly through my adobe sub. But I'm looking into investing into something. Thank you for this video!
Thanks for doing all the work to keep us all up to date Matt!
You're my go to source for all things AI
Thank you for this roundup, the side by side visual was a brilliant way to really see how they compare. I already subscribe to Leonardo (have to support my fellow Sydney siders) and might have to pay for ideogram soon too as it looks fantastic and gave me some excellent new variations on a project I am working on.
Midjourney always seems to prioritize generating cinematic, aesthetically pleasing images over prompt adherence. Which might be an advantage in some situations
that's funny, I found Firefly 2 one of the most realistic ones
not even the most sun hardened leather faced deep sea fisher actually look like those "realistic" photos
Agree love Dalles adherence. And we know we’ll see that with lead with Sora too. They’ll always be one to watch.
Glad to see more competition!!
Nice video! Love the banjo in the background
Firefly credits are included in your Adobe account. The amount of credits depends on the subscription to Adobe you have. Full suite 1000 credits or just Photoshop 250. These are also the credits used for gen fill. You get 1500 credits if you also have Adobe Stock, but that's dumb because of the cost and you can make the images instead.
I play around with a few webpage based models but at the end of the day I just boot up a local version of Stable Diffusion.
It does everything I need it to do for images.
Flux.1 is really good at prompt adherence, but requires a new prompting style, different from, for example, SD😊
Great comparisons! Just discovered this channel, good work reporting the development of AI image generator engines.
But the Flux 1.0 in Grok is Shnell model and it’s different than dev model. You should include dev in there as well
Matt, I can’t get Leonardo to create an old man looking at his younger self in a full length mirror. Ideally holding an electric guitar. I just get totally random responses.
3:01 a little bit noisy is actually good. These photographs look much more realistic with the film grain.
Agreed, noise isn't necessarily worse.
As long you put it into the prompt, it should be there. Otherwise it's bad
Thanks Boss BG ❤ so much work shared with us to help us and grow our knowledges 😁 yk I talk about your channel always 💯 for me , si far away, but weirdly so near, you are just the best 🔥🔥🔥
It’s funny when someone like Ideogram tries to work the word Idea into their name, but then you pronounce it like they wanted the word “Idiot” in their name 😂 Another excellent video about all the fun tools we have to play with 🎉
Midjourney needs to lower costs substantially, or they are going to be done for. Or at least give us a huge update that beats Magnific or something to justify it now.
Midjourney is done.
Good.Their customer service is horrible.
Let me tell you that free one is shit
😆 🤣 😂 no they're not........ but the stiff competition will hopefully make the devs stop screwing around with wasteful tasks such as developing unnecessary extra robust community interaction tools, etc, etc, etc.
If I want to interact with community...... that's what Discord and social media is for.
I just want the best AI image tool....... period.
@@bloodgourdthis. Garbage company
Nice dude.
Im not up to date with AI but its nice to see one actually get finished.
Thank you so much for doing this comparison! So interesting to see how they all compete with one another.
Great job breaking down current AI image generators in categories for comparisons.
Now that we have all of these free image generators, how long will it be before someone combines the best examples into a meta generation tool, which allows you to select the best images from five different models?
❗On Midjourney, to get a proper text rendering, it's better to put the word between "double quotes" rather than 'single quotes'
Thanks for your comparison Matt, this is really helpful!
Excellent run-down, thank you so much!! 😃
Wow that Black Forest Ham reference when talking about going Ham on Flux which is made by Black Forest Labs 💀😂🤣 I think it went over everyones head 😅 0:26
I found flux schnell the other day on Poe. Was so impressed I unsubscribed from MJ. I just could not ever get my words right to create what was on my mind in Midjourney. Flux understood me very well and it didn’t take hours of tweaking to get what I needed. ❤️🇦🇺
I think the Firefly fisherman is the best, TBH. All the others are too sharp/uncanny valley vibes.
It doesn't really follow the prompt though - where are the deep wrinkles? He doesn't seem particularly weathered, and the sea isn't that stormy either.
it would be neat if on the Figma board, you had all the test images for each model, so for example the 4 images for the old man sea photo for Ideogram arranged in a 2x2 grid, then users could zoom in to see the full quality of each
I've never cared too much about perfect AI text rendering cuz you get much finer control using photoshop, as long as there's a space for the text and a background that isn't too noisy, it's fine
Ideogram is way better at prompt adherence than Dall-E3.
Most text to image generators blow up if you ask for two distinct characters in the same prompt, and will often try to use them together (for example if you put Red Riding Hood and the Big Bad Wolf, they will give Red Riding hood fur).
But Ideogram does not have this issue, Flux is terrible at it, and Dall-E3 can do it but the context window prevents you from having specifics.
I use MJ for some crazy concept photos that could be used in Social Media posts, Ideogram and Flux for POD product.
Would be great if you give it a rating score like the way they do for softwares or hardwares
I love using Figma for narrowing down my iterations! Great rundown, thanks.
Thanks Mat this was great! Umm what was the uncensored version you mentioned last? 😅😅
The newer models seem to be limited to more generic styles, and trying a combination of named artists often is ignored. That’s one reason I’m still with MJ.
1.) Ideogram is the new king of the hill!
2.) Mystic has Flux underneath.
4.) Ideogram + Kling + Udio + Capcut 😂 = your personal Hollywood production company.
Very cool tutorial. Thank you so much for this. I'm a huge fan. I had a question about commercial rights. Do all them allow use commercially? I mainly use AI for print on demand so I would love to see examples or prompts for a flower pattern that's tiled, a cool sticker, a coloring book style design where it's black and white and a t-shirt design with text and a graphic with a black background so that I can remove it easily. Oh... I would also like to know the size. Are they all 1024x1024. Can I upscale them in app?
Thanks Matt... very helpful.
You should incorporate a infografic category, most of the models can't put details, text or another smart little elements or texts in to one single image, Ideogram 2.0 is the most impresive I have seen in doing that, and any other model can't even get close. Great to make a ppt
It would be great if you could do some testing of vector images as well.
Midjourney is still quite a long way ahead when prompting “in the style of”. The different styles and their aesthetic appeal is way beyond any other model including Flux.
That said I think their business model will need to change. Ideogram, Dalle and free huggingface Flux generations can do most things most people would need for free.
Would be great if there was an image "Battle" site like there is for LLMs.
I am a big fan of Playground and nightcafe and Leonardo you should try them too
Introducing Bro Ideogram 2.0, it's like a sudden burst of realism, providing highly realistic images
Thank you so much for this. It's exactly what I need 😇
It kind of seems like Midjourney Adobe and even runway prices are a little high. Hopefully they will make it a little more reasonable to compete with others. Competition can be good. I don’t think Midjourney free trial will help
Where is uncensored anything...how do you get it
This video is lit bro 💯💯💯💯💯💯
They don't offer free trials to new accounts anymore, they cancelled the free trial again
Leonardo ai is great and I prefer it over ideogram. While ideogram does the best with spelling of the words, but I think Leonardo produces much better images over all. As for Midjourney typography, it's disappointing that almost everything produced is misspelled and very small. A lot of the times on Midjourney the images show up behind characters or objects and this needs to be updated asap. Also I get frustrated with Midjourney's prompting, it's time they start making massive improvements. I like all three ai image generators and recommend each of them. Also they all have improved on the problems of hallucinations, specially on the hands.
I will never use Midjourney. They showed their true colors by being greedy while others were opensource and free. Now they want to backtrack and make it seem like they are doing a good thing, but they were just hindering artists trying to get their foot in the door with their paid options. Way better options now and I hope they get left behind in the dust.
You can generate all kinds of images on all kinds of platforms, ultimately the difference will be which users are profitable to these companies long-term and who will find uses for the work they create. It’s an expensive and very time-consuming hobby for the majority of users right now. How long people pursue something that requires so much personal resource without obvious financial return might ultimately be where AI image generators hit the buffers. Anything that becomes free and everyone can do will ultimately become devalued once the novelty goes.
“There isn’t enough time to lawyer up.”
This is why GROK 2 was released.
Heads up
Thx Matt ;
Jeremy
I guess what the user sees in the image matters a lot too. To me, the only one from the realistic/people category that actually looks like a real image is... Firefly. All the other, with the possible exception of Midjourney, are so painfully obviously fake that it hurts to look 😬 And I don't mean the images are bad, they're not, just that they scream AI generation
In addition, I have never got a good image from Leonardo's own models to the point that I've stopped even trying them out completely. Way, way too much contrast in every image, to me it spoils them completely. I'll reiterate that I don't mean the images are somehow universally bad, it's just so heavily dependent on the user. It's a blessing that we have many models to choose from, so everyone can decide for themselves!
AI Image generators are getting better. I will be coming back to this video, but I know it will probably be obselete in a year or maybe even a few months. Then it will be time for you to do a new version of this video which is great. lol
Ideogram is excellent. Have been messing around with it for a few days.
Matt Wolfe is now, without a shadow of a doubt, the go to guy for AI.
I noticed that for realism almost every model does a good job. But when you try to create scene with more than one faces, the difference becomes huge.
A-nother AI image generation video 👏
I have to disagree with Firefly. I think they have the most realistic image of the old man and sea (although you can't really see the sea). I wouldn't think its AI made ...
Asking the musical question… what does a sea of bubble tea really look like?
I think mid journey is really good at art styles, I’m also a pay user of Leonardo but it has difficulty of generate image with it’s style reference. Besides, I trained a model in Leonardo with my own photos and it just generate my “photo” whatever I try the prompts. Suggestions?
Great job! Thank you 😊
thanks Matt! 👍👍👍
I signed in with google and it still says I need a subscription to do anything or search for anything, didn't see anything about a free trial. Lots of vids and articles saying it's free but none with any useful information on how to do that, including this one.
Can you create video comparing ALL ai image generators side by side, and put them on figma like what you have did here? :) Thanks!
Now we need to build out an app like the Apple Store comparing models
i dont think midjourney can be beat when it comes to realism tbh, just looks way too good. glad I can use it for free again !!
Still nothing compares to Midjourney. A lot of people just don't know how to use it ;)
i like the fusion option best
Midjourney proves that excellent curation beats any tech. This is true in LLMs as well.
MJ is bad at prompt adherence but it produce the image so beautiful that made the user forget what their prompt was. The advantage of MJ is don't care about the prompt as long it produce the ultra beautiful image.