Dalle for photography images. Midjourney for illustration. I haven't tried Dalle (not received acceptance), but I do use Midjourney. I have noticed with MJ that there is a certain "sameness" to many of the compositions. But in all reality, we are just at the beginning of the AI revolution. It will get better.
Concur with all your points. Interestingly, I personally like very much the moody, nightmarish renders from Midjourney, which is hard to get from Dall-e. I have Dall-e but somehow found enjoying more the wacky colorful compositions midjourney gave me. Great video
I think this is solely personal taste. There were several points where you said "Dall-E is CLEARLY superior to Midjourney in this case" when it had generated simply an image, whereas Midjourney made an artwork or painting. For my needs & uses (mostly concept art), its tendency toward artistic and moody things is perfect and "CLEARLY superior". They're both good, but for very different uses I feel. (Example: generating a world map for D&D in Midjourney gives cool painted landscapes, whereas Dall-E 2 gave me... a road map.)
As you mentioned DALL-E and Midjourney are completely different systems. You can’t get the same results by typing the same basic prompts in both A.I’s. You can get realistic and astonishing results in Midjourney just by being more specific and detailed in your prompts.
I haven't used Dall-E but from what I've seen it does seem to have some advantages. In using Midjourney I quickly realised that aiming to produce something too specific will often see you sitting there redoing and tweaking prompts forever without ever getting what you want. If you can find the sweet spot of vague:specific and look for serendipitous results then it can produce good results. Your approach here was good although it still seems possible that understanding the language of midjourney and the technical ways to influence the results (--seeds, --style, --iw, text weights etc) is an important aspect and the tool hasn't had the chance to show what it can do if these tricks are implemented. Dall-E seems much easier to pick up but there is some potential that MJ could be a powerful tool and it's just the learning curve that is difficult (although it does appear that Dall-E has the edge anyway). One prompt trick I would have liked to see you try, is using art styles or artists to influence the style of the image. I have found that MJ works well with some styles and can emulate some more abstract artists very accurately.
Midjourney is all about your choice of "key" words in the prompt and also your settings, would recommend "--test --s 1250 --upbeta" / "--testp --s 1250 --upbeta" for better photo/portrait or use "normal or --hd" with the new "remaster" feature! This can make very good "painted" kinda photos as well, would still really like to get access to Dall•E 2 as well, mostly for the feature to replace objects or background
I think either can be like the other if you want. I wasn't impressed with Dall-E until I started adding Digital Art to everything. The stuff that comes out of midjourney with hyper-realistic is pretty close to Dall-E as well.
that's funny, I get the exact opposite result. For example, typing something simply like Cthulhu in Dall-e yields kindergarten drawings of something vaguely monstrous, while in midjourney I get exactly Cthulhu. as a tip adding keywords like "photograph" into midjourney, will yield something more akin to a photograph.
Your comment along with OP's video leads me to believe that some of my suspicions may be true that MidJourney has a really nefarious, malevolent tendency. I've OFTEN prompted very mundane, everyday things only to receive crazy nightmarish images! I am absolutely not into these and have had to alter my subject matter to reflect on MidJourney's penchant for dark images. I wonder, is this on purpose?
Good review. My experiences strongly mirror yours. As you said, MJ's strong suit is its painterly / creative interpretation of prompts, even as its current form is far from DAL's understanding of common language. My experiences have led me to see these as simply different engines. MJ's randomness and lack of realism is just as frustrating as DAL's lack of excitement and creativity. Both have strong pros and both have strong cons. The good news is that both engines are evolving quickly. Honestly, who knows where each of these engines will be in 6 months, much less than in a couple of years??
Um wtf? I find Midjourney bland with dall e there is different Pose and camera angle while all Midjourney has the same pose and camera angle. To get the best Dall e art type at the end" Digital art, Cinematic, Trending on artstation" and if u want Photorealistic Drawing type also "Photorealistic Illustration'. My profile pic is Dall e made
I honestly had a really hard time with both, I think maybe my concepts are just too abstract for them to understand, DALL-E seems to do well with those photo style renders but I was unable to get a usable face out of any of my tries and many of my digital art style images looked like a 3 year old drew them with crayons, that being said when I plugged in some things on sci-fi cities in digital art style it seemed to do really well… I like the textural quality and abstract art style of midjourney though it was also hit or miss I feel like my cpu time doesn’t go as far as DALL-E does, but I did end up with more usable images though none of them were clear either and I still had a really hard time getting it render results like what I see others do. The verdict is out for on whether it’s worth my money for a conceptual tool or not, I think I would probably just be able to draw what I want, though the color packets and textures are nice in both and I could use those as a starting, don’t know I’m not sure how valuable a tool this is for me right at this moment.
To get the photorealistic art type all this at end in Dall E 2 " Digital art, Photorealistic Illustration, Cinematic, Trending on artstation" All 4 and u can even add popular artist to copy they're art style like Leonardo Da Vinci etc
I really struggle to get good results with Dalle, whereas Midjourney is really killing it in my projects. I have had a lot of incredible results, even generating photos. With Dalle the results are pretty ...er... scary. I will say, that Midjourney really REALLY struggles generating anything other than moody color palettes. I require upbeat, sunny colors and lighting (in digital art) and getting those kinds of results are nearly impossible. Everything feels extremely heavy and dark or vintage. I am still trying to find the right prompts, so maybe it's just me. I feel like it's such a preference thing. As a creator I feel super comfortable with Midjourney and when I have to use Dalle I feel like it really just can't get me where I need to go so I probably don''t invest enough time trying to really get to know the program.
Thanks for sharing your test on prompts, what everyone should realize is that each platform has their own prompt key words and formatting. So to get the best results from your tools you have to educate yourself about each one because there is NO STANDARD (yet). I'm hoping to create my own similar video in response and I'll let you know when it's published. :)
Have you tried creating an image in one, then using that as a reference image in the other? Use Midjourney's abstract, painterly tendency to inform Dall-E's output.
Wow, this is an EXCELLENT VIDEO! I've had a GREAT deal of problems with MidJourney trying to get a specific object drawn traditionally (violin or guitar for example) No matter HOW many dozens of times I've altered the prompt, I get things that could POSSIBLY resemble a violin, yet nothing that was even close to usable. And I tried at LEAST 30-40 times. I'm definitely going to give Dalle 2 a try! Do they have a trial period? Thanks!
Thanks for the excellent "prompt teaching!" I'm definitely a Dall-E fan and will not go anywhere else (for now:-)! I began with Midjourny, but I will not go back for the reasons you mention. Even if I think there is an abundance of opportunities to redefine the prompts and get magnificent results in MidJourny, the Discord process is too messy for me:-)
great video, thanks for sharing : ) It would be nice to see another comparison focused on _long_ detailed prompts, to see which of the two can digest and successfully elaborate longer instructions... Dall-E2 seems to be more directable, but I've seen some spectacular midjourney results with long complex prompts...
Great suggestion! It's true Midjourney can produce some amazing images but when I look at these complex prompts I get the impression that it's a numbers game. Put everything in there and hope that something will stick! I read several of these complex prompts just to see how others built these images, and I've noticed that only a tiny fraction of that description is actually used in the final image. But I'm definitely adding this to the list of ideas. It'll make for a great video!
Depend on prompt you can create some stunning 3d realistic portrait at midjourney, you just need time to tweak it here and there. Words like photography, 3d, photorealistic, and use render engine that you prefer. I still prefer midjourney because for concept design midjourney could tickle your creative brain, and of course the pricing method why people still like midjourney. Overall I think this just early stage for public testing. I believe they will evole to something we cant imagine.
100% agree. We're still early in the game. I'm sure we'll see some amazing stuff in the coming months. Heck, I'm still blown away by some of the images produced with the current iterations of Midjourney and Dall•e 2!
For me, the result matters most, and if I have to tweak the description a few times to get what I want, that's not a problem. So having a system that adheres to that description and responds when I change it is best for me. Therefore, I agree with your general position that DALL-E is superior for most projects.
Adding 'happy' into the prompt might have helped (maybe you did and I missed that part). Either way, this is a really great help, each AI has its biases/use cases, like they all do and I don't want each program to be on par with each other! I wouldn't rate one above the other either, each has a different approach requiring different prompts to get similar results and I think that is a good thing! Each one has it's own specializations from what I see in this video and my own experience. God... this AI Art stuff is sooooooo interesting in ways I could write an entire Book on! Much of the times when I get totally different results than expected I've gotten some of my most favorite results! This is mostly using Stable Diffusion but it applies to all of this in the same way- I like the unexpected results, they give me ideas! I can see why it'd be undesirable if an Artist has a specific outcome expectation. But like I mentioned before, they each have benefits depending on how you're using it. I love the unexpected results, but that's due to how I use AI so far, I'm looking for broad results and the unexpected gives me inspiration a lot. Your Video has helped me with knowing how to apply both- MidJourney for unexpected things and inspiration, Dalle for refining things which I have a clear vision or end goal for. They both seem really great in their own way. Thank you for this video, it has helped a lot in many ways!
If you didn't get tired of videos about dalle, Piximperfect made one ("when AI edits your photos") showing how many things are possible within this platform. It would be interesting to watch the same with your own twist.
I just generated amazing sci-fi art in midjourney which looked more like unfinished concept art in dalle 2, I think they are both absolutely incredible tools for getting different things done
Nice to see other people doing what myself had to do sonner than later and save my time (plus old vocabulary remembered, I thought you said effy - girl's name of Greek origin meaning "fair speech", according to google - but it was iffy, a word I hadn't seen in decades)
It still amazes me that how AI can process Natural Languages and can understand it to make new images WOW ! we have come so far and who knows what will the future this is just beginning
Dall-E Understands short prompts better than Midjourney, tho, with Midjourney i managed to get very often better results, in an artistic, interesting way. I find mid journey way better with fantasy stuff, Dall-e better with realistic images. The biggest 2 problems with Dall-E are: 1) in order to get a good looking image you need to be very specific and know what you want to see, while in Mid every image looking so good! 2) when you trying to input lot of prompts into Dall-e you will get lot of bugs, stuff I can't even understand why the AI chose to show. Anyway, good video, but these are different AI, Dall-E is strong with few prompts so it's kind of unfair to Midjourney. If you test lot of prompts you'll see how Midjourney is better. Also I can assume you didn't used Midjourney many features like: --chaos, --style, weights::10 --testp (for photographic resaults), --quality, --ar Anyway, I just needed sort of justice for Midjourney xD
Pretty much everything came out of Midjourney are moody, 2d concept paintings; Dall-E on the other hand loves the stock photo looking images Stable is somewhere in the mid
Great comparison video. It would be interesting to see results If the prompt distinguished between art or photoreal for both systems. More over I'm curious if there are already equivalent ai that could generate videos, instead of just stills. That would be be awesome for B roll etc for editors.
Thank you for the video. Interesting. But I think it is not fair to compare the images directly from the same promts, because MidJourney is by defult more at the painted spectrum, while Dalle is more of the photo realism. To get a more fair comparison I think that you for example should give a promt in MidJourney to produce photo realism instead of painted, and THEN look at the different results. Have a beautiful day!
I agree MidJourney is difficult to direct with very specific prompts, but artistically it's so much better than dall e. Dall e almost always nails it in regards to prompt but images often look like they were made by an amateur artist.
I'm finding that unless you specify a male subject, they will always choose female first. I imagine it could be a result of there simply being so many more images of women online to train the AIs with.
I keep hoping to hear about a text to 3D diffuser or something that works along those lines. Being able to dream up amazing things and then have them available as 3D models. It would be truly magical. My main love is 3D printing and I don't have to tell anyone reading this just how stunning A.I generated real world models would be. Heck why not dream further and have the engine intelligently cut up the model for any printer size and add the smartest supports too. A guy can dream as big as these A.I's
Tried the watch in a hand and got 1 out of 20 pictures that where good. Stable diffusion with SD2.1 model. You can generate this pics quite fast with SD but doing this with Midjourney would take forever. Midjourney and Stable diffusion using the same tech. I did. try the red bal on a wooden table and it worked great but „red ball on wooden table standing on green carpet“ failed
I always get awkward results with DALL-E, especially distorted faces on human and animals, way to far from the photorealism I see in this video. Is there an upgrade for obtaining quality like this?
The systems are different. Prompt engineering is needed because the systems detect information differently. In my opinion, Dall-E 2 is inferior to Midjourney-V4 when it comes to many aspects; creativity, stylized creations, artistic expression, color sense, angles, and general composition. However, Dall-E 2 has Midjourney beat when it comes to more photo-realistic images, and the customization that is offered with the final product. However, MidJourney--v4 offers final product customization, albeit still not as great as Dall-E 2. Heres how the prompts are different. If I were to try and get a similar styled piece with each I would have to alter the prompts quite extensively: Midjourney: "a portrait of a young girl with dirty blonde hair and bright blue eyes sitting on a mossy rock in the forest, hyperrealistic, wide angle lens, lens flare, bokeh background, lens blur on background, lens focus on face, 4K, detailed, gorgeous, sharp shadows, photorealism, natural lighting, natural colors, -V4 " Dall-E 2: "a portrait of a young girl with dirty blonde hair and bright blue eyes sitting on a mossy rock in the forest, blurred background with bokeh, sharp shadows, wide angle lens" that's the difference. Dall-E 2 automatically tries to go for a hyperrealistic look while Midjourney automatically tries to go for a highly stylized, expressionistic look. Midjourney might require you to type more but in the end, the results are better than Dall-E 2, but in both you have to get skilled at learning how the A.I detects information and how you should best prompt it to get what you want.
I have been using midjourney, and it seems the more I learn about prompts, the worse my photos become. When I just enter 1 word descriptions, the photos are incredible but if I put a lot of descriptive words on what I want, the less it becomes what I want. I will try dalle e
I agree. Midjourney even mentions in the documentation that the more abstract the description the better the images will be. But I also think that the more descriptive prompts suffer from the fact that you’ve already formed the images in your head. Without any expectations ( the simple prompts) the results will automatically be much better! Definitely give Dalle a try. I think you will like it quite a bit.
I typed "a male lion roaming on the African plains" in mid journey I got a pile of straw in Dall e I did get a male lion on the African plains. But no artistic merit. I tweeted it with Absract painting and got the image I wanted. But with mid journey I get great landscapes. And fantasy pictures. So if I don't get what I want from one I used the other
I’ve been using DALL-E 2 for just a while. I started with a project where I wanted things to look the same, so I stuck to a single art style, and have not yet done very much exploration of how it produces other art styles. And I have, of course, run into the same problems with weirdly distorted outcomes, or what I call a Star Trek transporter accident, where someone has three arms or things growing out of their head! But when I see people talking about MidJourney and they scroll through images, the MidJourney stuff looks way better than anything I’ve seen from DALL-E 2. But this side by side comparison shows that it does have its problems. I might want that “MidJourney look” for some things, but not want it for other things. This could get expensive! Of course, you chose to feature three things that all of these AI art generators have trouble with; faces, hands and text. But that tiny house comparison was revealing, DALL-E2 did much better. And MidJourney’s inability to be photographic with those female faces, but instead be painterly was revealing. And those mutated hands with the watches were much worse than what DALL-E did with its hands. But… I think I liked the MidJourney cityscape comparison better. And the first sad girl images from MidJourney. Oh well, onward in perfecting my DALL-E 2 prompt writing skills! Maybe do another video making comparisons that don’t feature faces, hands and text.
PLEASE, DALL-E2 is good for photography, but MIDJOURNEY it's so fabulous, for creating the spectacular Image, for me MIDJOURNEY IS the best really really
I liked the Video! I would care about a video where you don't use the same promt but go for a specific outcome. To me both AI are still very good for inspiration and happy little accidents but hard for specific stuff you have in mind. Especially Fantasy Illustration where a established world with a distinctive look to fantasy creatures are established. I agree in the assessment that DALL-E is easier to work with since it "listens" better. Midjourny is easier in my opinion if you just want something nice "ready to use"
the only detail is that you will always most likely start with the same idea of a prompt, regardless of the engine. And you don't always have enough credits to reach a specific outcome
I tried both AI today, and i found MidJourney is far better than DALLE in almost every case. You need to update your opinion in the video. Things changed a lot during the past few months.
Funny how I'm trying the very same prompts in Dalle2 and I get some awful disfigured faces or bad shadow and light scenes... 😢 I'm trying to get a sad big bodied woman sitting on a chair with the same layout you had in the first comparison... No success yet 😔
You saved me a lot of time trying to experiment with Midjourney as another tool, which I just find "dark" and not "elegant" or "special" from your examples. I am nowhere near close to exhausting the various artist and painterly "styles" that Dall-E gives me, including the incredible Outpainting. I use Klimt as my favorite artist to mimic, and I have managed to coax giant wall art from various such artists with Dall-E outputs that I would be proud to hang on entire home, apartment, or gallery walls, as I specialize in 4'x8' panels... now HOW to get that size imaged and printed on stretch canvas or fabric? That I don't know yet, I might have to use projectors to throw the digital images onto the walls or panels... and then I probably need to deepen the resolution, and there are tools for that.
to use the same prompts is not really good for comparision. for example if u add --testp to midjourney it will create a photo like image and without it will create a hand drawn or cgi look.
Midjourney actually distinguishes between "artwork" or "Painting" and "photograph". Dimitris, you need to tell Midjourney that you WANT a photograph not a painting. This is an unfair comparison. Also some only want painting and not photographs as they find photographs boring. After all MJ and Dalle are supposed to be ART tools not photography tools!
Pour moi, Midjourney et Dall-e sont complémentaires. J'utilise les 2. Midjourney offre un résultat plus artistique, plus graphique que Dall-e et ce dernier offre un résultat plus photographique.
first prompt : "beautifull woman" -> dalle spits out horse faced womens while midjourney really something which resembles a beautifull womon -> Dimitris Katsafouros dalle is better ... i would not take anything what this guy says at face value lol
At the beginning of video, you said you've got no horse in the game.. After watching most of the video... Seems to me your riding that dalle2 horse over the finishing line.
From a person that has used MJ. I think you better check your settings. I have never had a single image look as bad as what this guy gets over and over again.
Dalle for photography images. Midjourney for illustration.
I haven't tried Dalle (not received acceptance), but I do use Midjourney. I have noticed with MJ that there is a certain "sameness" to many of the compositions. But in all reality, we are just at the beginning of the AI revolution. It will get better.
better but not for us :0
@@SAMAprishla no, MidJourney does better jobs on both than DALLE
Concur with all your points. Interestingly, I personally like very much the moody, nightmarish renders from Midjourney, which is hard to get from Dall-e. I have Dall-e but somehow found enjoying more the wacky colorful compositions midjourney gave me. Great video
I think this is solely personal taste. There were several points where you said "Dall-E is CLEARLY superior to Midjourney in this case" when it had generated simply an image, whereas Midjourney made an artwork or painting. For my needs & uses (mostly concept art), its tendency toward artistic and moody things is perfect and "CLEARLY superior". They're both good, but for very different uses I feel.
(Example: generating a world map for D&D in Midjourney gives cool painted landscapes, whereas Dall-E 2 gave me... a road map.)
Agreed. With most of these examples, Midjourney's results were much more to my liking. The video is very informative though.
As you mentioned DALL-E and Midjourney are completely different systems. You can’t get the same results by typing the same basic prompts in both A.I’s. You can get realistic and astonishing results in Midjourney just by being more specific and detailed in your prompts.
I disagree. Please see my post above/below if interested in discussion.
i totally agree. it is in the prompting. this guy do not understand how to use midjourney.
I haven't used Dall-E but from what I've seen it does seem to have some advantages. In using Midjourney I quickly realised that aiming to produce something too specific will often see you sitting there redoing and tweaking prompts forever without ever getting what you want. If you can find the sweet spot of vague:specific and look for serendipitous results then it can produce good results.
Your approach here was good although it still seems possible that understanding the language of midjourney and the technical ways to influence the results (--seeds, --style, --iw, text weights etc) is an important aspect and the tool hasn't had the chance to show what it can do if these tricks are implemented. Dall-E seems much easier to pick up but there is some potential that MJ could be a powerful tool and it's just the learning curve that is difficult (although it does appear that Dall-E has the edge anyway).
One prompt trick I would have liked to see you try, is using art styles or artists to influence the style of the image. I have found that MJ works well with some styles and can emulate some more abstract artists very accurately.
Midjourney is all about your choice of "key" words in the prompt and also your settings, would recommend "--test --s 1250 --upbeta" / "--testp --s 1250 --upbeta" for better photo/portrait or use "normal or --hd" with the new "remaster" feature! This can make very good "painted" kinda photos as well, would still really like to get access to Dall•E 2 as well, mostly for the feature to replace objects or background
What does 'upbeta' do?
I think either can be like the other if you want.
I wasn't impressed with Dall-E until I started adding Digital Art to everything. The stuff that comes out of midjourney with hyper-realistic is pretty close to Dall-E as well.
that's funny, I get the exact opposite result. For example, typing something simply like Cthulhu in Dall-e yields kindergarten drawings of something vaguely monstrous, while in midjourney I get exactly Cthulhu.
as a tip adding keywords like "photograph" into midjourney, will yield something more akin to a photograph.
That is because MidJourney got updated with a new algorithm recently
@@Azolture right but this is a new video
Your comment along with OP's video leads me to believe that some of my suspicions may be true that MidJourney has a really nefarious, malevolent tendency. I've OFTEN prompted very mundane, everyday things only to receive crazy nightmarish images! I am absolutely not into these and have had to alter my subject matter to reflect on MidJourney's penchant for dark images. I wonder, is this on purpose?
@@skyko I'm not sure. You should join the midjourney discord. They have office hours where the developers answer questions.
And also, try the --chaos flag on Midjourney. There are so many more options hidden away which make it way more powerful.
I prefer midjourney. I don't want a result I already have in my head, I want inspiration and surprise.
Not sure if you did this or not, but you need to do the same test again with Midjourney version 4. This new version is unbelievable.
Good review. My experiences strongly mirror yours. As you said, MJ's strong suit is its painterly / creative interpretation of prompts, even as its current form is far from DAL's understanding of common language. My experiences have led me to see these as simply different engines. MJ's randomness and lack of realism is just as frustrating as DAL's lack of excitement and creativity. Both have strong pros and both have strong cons. The good news is that both engines are evolving quickly. Honestly, who knows where each of these engines will be in 6 months, much less than in a couple of years??
Um wtf? I find Midjourney bland with dall e there is different Pose and camera angle while all Midjourney has the same pose and camera angle.
To get the best Dall e art type at the end" Digital art, Cinematic, Trending on artstation" and if u want Photorealistic Drawing type also "Photorealistic Illustration'.
My profile pic is Dall e made
Sometimes we end up with both companies combining into one, or like VHS vs Beta, one 'wins'.
DALL-E is best for realistic photos. Midjourney for stylized photos.
I honestly had a really hard time with both, I think maybe my concepts are just too abstract for them to understand, DALL-E seems to do well with those photo style renders but I was unable to get a usable face out of any of my tries and many of my digital art style images looked like a 3 year old drew them with crayons, that being said when I plugged in some things on sci-fi cities in digital art style it seemed to do really well… I like the textural quality and abstract art style of midjourney though it was also hit or miss I feel like my cpu time doesn’t go as far as DALL-E does, but I did end up with more usable images though none of them were clear either and I still had a really hard time getting it render results like what I see others do. The verdict is out for on whether it’s worth my money for a conceptual tool or not, I think I would probably just be able to draw what I want, though the color packets and textures are nice in both and I could use those as a starting, don’t know I’m not sure how valuable a tool this is for me right at this moment.
To get the photorealistic art type all this at end in Dall E 2 " Digital art, Photorealistic Illustration, Cinematic, Trending on artstation" All 4 and u can even add popular artist to copy they're art style like Leonardo Da Vinci etc
I really struggle to get good results with Dalle, whereas Midjourney is really killing it in my projects. I have had a lot of incredible results, even generating photos. With Dalle the results are pretty ...er... scary. I will say, that Midjourney really REALLY struggles generating anything other than moody color palettes. I require upbeat, sunny colors and lighting (in digital art) and getting those kinds of results are nearly impossible. Everything feels extremely heavy and dark or vintage. I am still trying to find the right prompts, so maybe it's just me. I feel like it's such a preference thing. As a creator I feel super comfortable with Midjourney and when I have to use Dalle I feel like it really just can't get me where I need to go so I probably don''t invest enough time trying to really get to know the program.
Thanks for sharing your test on prompts, what everyone should realize is that each platform has their own prompt key words and formatting. So to get the best results from your tools you have to educate yourself about each one because there is NO STANDARD (yet). I'm hoping to create my own similar video in response and I'll let you know when it's published. :)
Have you tried creating an image in one, then using that as a reference image in the other? Use Midjourney's abstract, painterly tendency to inform Dall-E's output.
Nice work. I have gotten usable images from both. Great job highlighting the differences.
Wow, this is an EXCELLENT VIDEO! I've had a GREAT deal of problems with MidJourney trying to get a specific object drawn traditionally (violin or guitar for example) No matter HOW many dozens of times I've altered the prompt, I get things that could POSSIBLY resemble a violin, yet nothing that was even close to usable. And I tried at LEAST 30-40 times. I'm definitely going to give Dalle 2 a try! Do they have a trial period? Thanks!
I don't believe Dalle has trials but it's super affordable. You can just add 10 euros or so and it will last you for lots of prompts.
@@marvelousdecay Thank you for the info, I will def check it out! 😉
Thanks for the excellent "prompt teaching!" I'm definitely a Dall-E fan and will not go anywhere else (for now:-)! I began with Midjourny, but I will not go back for the reasons you mention. Even if I think there is an abundance of opportunities to redefine the prompts and get magnificent results in MidJourny, the Discord process is too messy for me:-)
Would be interesting to see DallE vs Midjourney's features --testp and --test
i was thinking that the whole time
great video, thanks for sharing : )
It would be nice to see another comparison focused on _long_ detailed prompts, to see which of the two can digest and successfully elaborate longer instructions... Dall-E2 seems to be more directable, but I've seen some spectacular midjourney results with long complex prompts...
Great suggestion!
It's true Midjourney can produce some amazing images but when I look at these complex prompts I get the impression that it's a numbers game. Put everything in there and hope that something will stick!
I read several of these complex prompts just to see how others built these images, and I've noticed that only a tiny fraction of that description is actually used in the final image.
But I'm definitely adding this to the list of ideas. It'll make for a great video!
Depend on prompt you can create some stunning 3d realistic portrait at midjourney, you just need time to tweak it here and there. Words like photography, 3d, photorealistic, and use render engine that you prefer. I still prefer midjourney because for concept design midjourney could tickle your creative brain, and of course the pricing method why people still like midjourney. Overall I think this just early stage for public testing. I believe they will evole to something we cant imagine.
100% agree. We're still early in the game. I'm sure we'll see some amazing stuff in the coming months. Heck, I'm still blown away by some of the images produced with the current iterations of Midjourney and Dall•e 2!
I've always wondered--will two people inserting two same prompts get images that are the same? Assuming they're using the same system
Hi Dimitris great comparison! Mind giving Bluewillow a try and differentiate the results vs hyped AIs just to see how they have progressed?
For me, the result matters most, and if I have to tweak the description a few times to get what I want, that's not a problem. So having a system that adheres to that description and responds when I change it is best for me. Therefore, I agree with your general position that DALL-E is superior for most projects.
Adding 'happy' into the prompt might have helped (maybe you did and I missed that part). Either way, this is a really great help, each AI has its biases/use cases, like they all do and I don't want each program to be on par with each other! I wouldn't rate one above the other either, each has a different approach requiring different prompts to get similar results and I think that is a good thing! Each one has it's own specializations from what I see in this video and my own experience. God... this AI Art stuff is sooooooo interesting in ways I could write an entire Book on!
Much of the times when I get totally different results than expected I've gotten some of my most favorite results! This is mostly using Stable Diffusion but it applies to all of this in the same way- I like the unexpected results, they give me ideas! I can see why it'd be undesirable if an Artist has a specific outcome expectation. But like I mentioned before, they each have benefits depending on how you're using it. I love the unexpected results, but that's due to how I use AI so far, I'm looking for broad results and the unexpected gives me inspiration a lot.
Your Video has helped me with knowing how to apply both- MidJourney for unexpected things and inspiration, Dalle for refining things which I have a clear vision or end goal for. They both seem really great in their own way. Thank you for this video, it has helped a lot in many ways!
If you didn't get tired of videos about dalle, Piximperfect made one ("when AI edits your photos") showing how many things are possible within this platform. It would be interesting to watch the same with your own twist.
I just generated amazing sci-fi art in midjourney which looked more like unfinished concept art in dalle 2, I think they are both absolutely incredible tools for getting different things done
Nice to see other people doing what myself had to do sonner than later and save my time (plus old vocabulary remembered, I thought you said effy - girl's name of Greek origin meaning "fair speech", according to google - but it was iffy, a word I hadn't seen in decades)
It still amazes me that how AI can process Natural Languages and can understand it to make new images WOW ! we have come so far and who knows what will the future this is just beginning
Dall-E Understands short prompts better than Midjourney, tho, with Midjourney i managed to get very often better results, in an artistic, interesting way.
I find mid journey way better with fantasy stuff, Dall-e better with realistic images.
The biggest 2 problems with Dall-E are:
1) in order to get a good looking image you need to be very specific and know what you want to see, while in Mid every image looking so good!
2) when you trying to input lot of prompts into Dall-e you will get lot of bugs, stuff I can't even understand why the AI chose to show.
Anyway, good video, but these are different AI, Dall-E is strong with few prompts so it's kind of unfair to Midjourney.
If you test lot of prompts you'll see how Midjourney is better.
Also I can assume you didn't used Midjourney many features like:
--chaos, --style, weights::10 --testp (for photographic resaults), --quality, --ar
Anyway, I just needed sort of justice for Midjourney xD
Not me. I find Midjourney prompt bland with the same Pose while Dall e 2 has different kind of pose and camera angle
Pretty much everything came out of Midjourney are moody, 2d concept paintings; Dall-E on the other hand loves the stock photo looking images
Stable is somewhere in the mid
Great comparison video.
It would be interesting to see results If the prompt distinguished between art or photoreal for both systems.
More over I'm curious if there are already equivalent ai that could generate videos, instead of just stills. That would be be awesome for B roll etc for editors.
Not aware of an AI for videos but that would be wild!
I'm sure this would introduce a whole host of issues the developers would have to tackle!
@@marvelousdecay staple diffusion you should compare staple diffusion with daile 2 and mid journey
Thank you for the video. Interesting.
But I think it is not fair to compare the images directly from the same promts, because MidJourney is by defult more at the painted spectrum, while Dalle is more of the photo realism.
To get a more fair comparison I think that you for example should give a promt in MidJourney to produce photo realism instead of painted, and THEN look at the different results.
Have a beautiful day!
I agree MidJourney is difficult to direct with very specific prompts, but artistically it's so much better than dall e. Dall e almost always nails it in regards to prompt but images often look like they were made by an amateur artist.
I'm finding that unless you specify a male subject, they will always choose female first. I imagine it could be a result of there simply being so many more images of women online to train the AIs with.
I keep hoping to hear about a text to 3D diffuser or something that works along those lines. Being able to dream up amazing things and then have them available as 3D models. It would be truly magical. My main love is 3D printing and I don't have to tell anyone reading this just how stunning A.I generated real world models would be. Heck why not dream further and have the engine intelligently cut up the model for any printer size and add the smartest supports too. A guy can dream as big as these A.I's
4:39 the one on the right, even though it didn't follow the prompt, I think that one is a neat painting
Why is the face generated by Midjourney completely different from the reference photo, please tell me how to do it?
Great looking video… how did you put the images in moving frames in the intro?
what a good idea.
So we can compare the style .
Tried the watch in a hand and got 1 out of 20 pictures that where good. Stable diffusion with SD2.1 model.
You can generate this pics quite fast with SD but doing this with Midjourney would take forever.
Midjourney and Stable diffusion using the same tech. I did. try the red bal on a wooden table and it worked great but „red ball on wooden table standing on green carpet“ failed
I always get awkward results with DALL-E, especially distorted faces on human and animals, way to far from the photorealism I see in this video. Is there an upgrade for obtaining quality like this?
The systems are different.
Prompt engineering is needed because the systems detect information differently. In my opinion, Dall-E 2 is inferior to Midjourney-V4 when it comes to many aspects; creativity, stylized creations, artistic expression, color sense, angles, and general composition. However, Dall-E 2 has Midjourney beat when it comes to more photo-realistic images, and the customization that is offered with the final product. However, MidJourney--v4 offers final product customization, albeit still not as great as Dall-E 2.
Heres how the prompts are different. If I were to try and get a similar styled piece with each I would have to alter the prompts quite extensively:
Midjourney: "a portrait of a young girl with dirty blonde hair and bright blue eyes sitting on a mossy rock in the forest, hyperrealistic, wide angle lens, lens flare, bokeh background, lens blur on background, lens focus on face, 4K, detailed, gorgeous, sharp shadows, photorealism, natural lighting, natural colors, -V4 "
Dall-E 2: "a portrait of a young girl with dirty blonde hair and bright blue eyes sitting on a mossy rock in the forest, blurred background with bokeh, sharp shadows, wide angle lens"
that's the difference. Dall-E 2 automatically tries to go for a hyperrealistic look while Midjourney automatically tries to go for a highly stylized, expressionistic look. Midjourney might require you to type more but in the end, the results are better than Dall-E 2, but in both you have to get skilled at learning how the A.I detects information and how you should best prompt it to get what you want.
I have been using midjourney, and it seems the more I learn about prompts, the worse my photos become. When I just enter 1 word descriptions, the photos are incredible but if I put a lot of descriptive words on what I want, the less it becomes what I want. I will try dalle e
I agree. Midjourney even mentions in the documentation that the more abstract the description the better the images will be.
But I also think that the more descriptive prompts suffer from the fact that you’ve already formed the images in your head.
Without any expectations ( the simple prompts) the results will automatically be much better!
Definitely give Dalle a try. I think you will like it quite a bit.
Easy summary:
Midjourney: Artistic
Dall.E2: Photographic
Mid journey was available instantly where as dall e is making me wait weeks now to access it. So guess I know which one I'm going to use.
Why does delle create photo type images? Does it do anything else like drawings paintings animation, etc?
Yes
I typed "a male lion roaming on the African plains" in mid journey I got a pile of straw in Dall e I did get a male lion on the African plains. But no artistic merit. I tweeted it with Absract painting and got the image I wanted. But with mid journey I get great landscapes. And fantasy pictures. So if I don't get what I want from one I used the other
I’ve been using DALL-E 2 for just a while. I started with a project where I wanted things to look the same, so I stuck to a single art style, and have not yet done very much exploration of how it produces other art styles. And I have, of course, run into the same problems with weirdly distorted outcomes, or what I call a Star Trek transporter accident, where someone has three arms or things growing out of their head! But when I see people talking about MidJourney and they scroll through images, the MidJourney stuff looks way better than anything I’ve seen from DALL-E 2. But this side by side comparison shows that it does have its problems. I might want that “MidJourney look” for some things, but not want it for other things. This could get expensive! Of course, you chose to feature three things that all of these AI art generators have trouble with; faces, hands and text. But that tiny house comparison was revealing, DALL-E2 did much better. And MidJourney’s inability to be photographic with those female faces, but instead be painterly was revealing. And those mutated hands with the watches were much worse than what DALL-E did with its hands. But… I think I liked the MidJourney cityscape comparison better. And the first sad girl images from MidJourney. Oh well, onward in perfecting my DALL-E 2 prompt writing skills! Maybe do another video making comparisons that don’t feature faces, hands and text.
Where can i try dall-e 2?
PLEASE, DALL-E2 is good for photography, but MIDJOURNEY it's so fabulous, for creating the spectacular Image, for me MIDJOURNEY IS the best really really
What about a comparison between DALL-E & Diffusion?
Wow, thanks for a fantastic vid!
Midjourny need some extra details about the prompt and you can do heavy customisation that why it's best in my opinion
This needs to be upddated taken into consideration the more realistic test on midjourney
Can't wait for the close future where we have 30 or more different AIs and we look gigantic comparisons between them!
hehe. A.I overload! :D
Sure then you can PAY 30 subscriptions to use them and pay just to be their Tester !!!
4:49 which is exactly why i dig Midjourney more than Dall-E-2.
This doesn’t mean that Dall-E-2 can’t make moody images?
I liked the Video!
I would care about a video where you don't use the same promt but go for a specific outcome.
To me both AI are still very good for inspiration and happy little accidents but hard for specific stuff you have in mind. Especially Fantasy Illustration where a established world with a distinctive look to fantasy creatures are established. I agree in the assessment that DALL-E is easier to work with since it "listens" better. Midjourny is easier in my opinion if you just want something nice "ready to use"
the only detail is that you will always most likely start with the same idea of a prompt, regardless of the engine. And you don't always have enough credits to reach a specific outcome
Midjourney is the Hieronymus Bosch of AI art generators.
have you done an updated video
I tried both AI today, and i found MidJourney is far better than DALLE in almost every case. You need to update your opinion in the video. Things changed a lot during the past few months.
Loving it man. Digital crack. Great work.
i have question. is dall-e 2 and dall-e is same? and how to get acces to dalle-2 when you already have dall-e
It's not the same. You need to apply for dall e 2 access and wait for email.
Well since I have never been able to use Dalle 2 I vote for Midjourney.
you gotta specify you want it to be a photograph, midjourney is so good if you know how to use it
Dimitris is talking like an AI but its cool.
Funny how I'm trying the very same prompts in Dalle2 and I get some awful disfigured faces or bad shadow and light scenes... 😢 I'm trying to get a sad big bodied woman sitting on a chair with the same layout you had in the first comparison... No success yet 😔
You saved me a lot of time trying to experiment with Midjourney as another tool, which I just find "dark" and not "elegant" or "special" from your examples. I am nowhere near close to exhausting the various artist and painterly "styles" that Dall-E gives me, including the incredible Outpainting. I use Klimt as my favorite artist to mimic, and I have managed to coax giant wall art from various such artists with Dall-E outputs that I would be proud to hang on entire home, apartment, or gallery walls, as I specialize in 4'x8' panels... now HOW to get that size imaged and printed on stretch canvas or fabric? That I don't know yet, I might have to use projectors to throw the digital images onto the walls or panels... and then I probably need to deepen the resolution, and there are tools for that.
Jesus that's depressing. You're going to drape the walls in fraudulent author-less machine generated Klimt knock offs.
Both, as they are that opposite. Different tools, more and fast produtivity.
I like Dall-e better. Midjourney would be perfect at making a heavy metal album cover.
Not a fair test MJ interprets portrait to mean a drawing or painting,you need to begin your prompt with the word photo
need to redo this video midjourny has had an amazing update! fully life like 8k renders! now!!
to use the same prompts is not really good for comparision. for example if u add --testp to midjourney it will create a photo like image and without it will create a hand drawn or cgi look.
I tried Dall-E same commands and was completly disapointed. No interesting pictures, but truth is, more realistic.
Midjourney actually distinguishes between "artwork" or "Painting" and "photograph". Dimitris, you need to tell Midjourney that you WANT a photograph not a painting. This is an unfair comparison. Also some only want painting and not photographs as they find photographs boring. After all MJ and Dalle are supposed to be ART tools not photography tools!
Hopefully they won't be able to make the AI do 3D Motion Design videos for quite some time.
I do a lot of photo realistic work so dall e works better for me and also MJ interface is a bit of a nightmare to deal with. 😅
how to get access to DALL-E-2?
Apply to their beta and wait! Unfortunately that's the only way so far!
Pour moi, Midjourney et Dall-e sont complémentaires. J'utilise les 2. Midjourney offre un résultat plus artistique, plus graphique que Dall-e et ce dernier offre un résultat plus photographique.
this must be before the update in midjourney --test --testp
We’re all art directors know.
Irrelevant today, just for the photos, you did not test Midjourney with the option --testp
I made better real portrait photography with midjourney than dalle using upbeta and testp prompts.
Well I can't access Dall-E unfortunately pffft
great video
Dalle 2 is more realistic meanwhile midjourney feels more artistic.
It seems to me like Midjourney is more Dali-like than Dall-E…
first prompt : "beautifull woman" -> dalle spits out horse faced womens while midjourney really something which resembles a beautifull womon -> Dimitris Katsafouros dalle is better ... i would not take anything what this guy says at face value lol
W O W, absolute gorgeous! Masterpieces. Love AI
3 months later Midjourney do a really different results ... with AI any kind of comparison like this is outdated after couple weeks :D
At the beginning of video, you said you've got no horse in the game.. After watching most of the video... Seems to me your riding that dalle2 horse over the finishing line.
im confused im a few mins in and u dont understand how to direct it if u dont want a head on say the angle shot u like
be specefic
From a person that has used MJ. I think you better check your settings. I have never had a single image look as bad as what this guy gets over and over again.
Check the date of the video. The results you see in the video resemble the developments of the program at that time.
make sure to add --v 4
4:43 agree, mine too 😂😂
😆
DOLL E (DAHL EE). like the artist Salvador Dali.
Couldn't have disagree more lol, Dall-E to me seems like just another photography website out there, I kinda get it but not really lol
Before you compare, you have to improve a lot on Midjourney because there is a curse of progression. Your current level is still too low.
This is the worst comparation ever between dalle an midjourney. You dont know how to use midjourney
I prefer Dalle, brah