MidJourney does have negative prompts using “-no” (V4 didn’t at first but does now). It also has weighted prompts, image prompts, “remix” (changing the prompt as you choose variations). It does not have in/out painting though.
Stable Diffusion is a do-it-all locally with lots of plug-ins and API's... it's like comparing a 99 in one Swiss army knife to 2 regular knifes of same size and saying it's about the same. oh, and the multi-purpose one is totally free and don't depend on corporate agenda and connectivity :)
I love freedom software, and free software. But for people who _aren't_ broke, maybe… just maybe… they would rather spend a little money than spend a little time convincing SD not to generate nightmare fuel? Y'know, like it did all over this competition?
Yeah unfortunately you’d have to use the prompts that best fit’s each AI. This is like comparing program languages using the syntax of one and hoping for good results for the others. You must use negatives in SD2 = whole new ball game. As far as more intuitive for the general public then I’d have to give it to MJ. But to literally get amazing works in any of them you’re going to have to use the language prompts the fit with that particular AI.
SD2 is extremely reliant on negative prompts to yield good results. Without them it's generally worse than 1.5. As long as you specifically ask it to not make anything ugly it's actually really good.
I’d be interested to see how this works out when using stable diffusion with negative prompts since this is the correct way to use it. Also, a test with F222 included for comparison would be good
I tried a bunch of ai for art and so far midjourney is simply on another level, never been this amazed and lowkey horrified at how fast a computer can give u something so accurate, detailed and stunning.
Midjourney was on point in almost all categories, stable diffusion also gave some interesting results especially in the design category. Dall-e seams to be already behind the competition. Almost never made something better than the other two.
MJ V5 is aiming to be out by Jan and they said hands will be improved. V4 still has issues but definitely much better than it used to be. Often the mistakes are more subtle now.
Midjourney appears to have the best results in this video but as others have pointed out, these are not at all what properly prompted Stable Diffusion 2 images look like. I've used all 3 Ai shown here and this is heavily biased because of the prompts. I doubt that the producer was intentionally doing that, and is probably just a lack of knowledge. That said, Midjourney is indeed great... if you don't mind having to pay for every image you produce, which is something that should also be mentioned. Stable Diffusion is completely free if you have the hardware to run it locally. There is no free option for Midjourney.
Hey Oddar, thanks for watching and commenting on the video! The idea behind this video was to have some fun and see how different models would react to the same prompts. These prompts are not too simple or customized for each model - in fact, they are typically used in stable diffusion. As it turns out, midjourney responds well to simple prompts, as shown in the video. We thought about making a video to show which prompts work best for each model (such as a negative prompt for SD2, for example), but there are probably already plenty of videos out there that explain all that stuff much better than we could. Thank you again for watching and commenting, and I hope you are subscribed and have the notification bell turned on
@@nobodyandthecomputer I did enjoy your video. I was just pointing out that without the correct negative prompts for SD 2 it can suffer quite visibly in image quality. This is far more important for SD 2 than it was for SD 1. People who aren't aware of this will judge those images poorly, when in fact with a generic negative prompt the improvement can be extreme. I quite like Midjourney and think it may indeed have an edge over Stable Diffusion when comparing the base models, but that changes when you use alternate SD models which is something that Midjourney can't do... although that is definitely outside the scope for a generalized video like yours,
I was expecting MJ to win hands down but was really impressed with Dall-e's output. I'm wondering if a lot of people are having too much personal bias towards MJ to be more objective. Given that SD is generally bleh compared to them, a few times it came out impressive.
Stable diffusion 2 sadly suffers from need to use negative prompts to work, the backlash made the guys from SD to release SD 2.1 which works better without using as much negative prompts. Don't know if you used amply negative prompts in this test. No doubt MJ v4 is extremely potent, the best overall for now, nevertheless.
As fun as this is, it's not a fair comparison in the least. I love me some mid journey but she is limited on features. SDXL has a ton of extra doodads that you can refine an image with to make something absolutely perfect. Going with the first image isn't really showing the strengths or limitations of each Platform
it's sad that only stable diffusion can be run natively on a personal computer, but luckily you can actually train stable diffusion to produce midjourney style images lol
In truth, SD has no competition for several reasons - first, it can be used locally, it is supported by a community that trains models for it itself, and it has convenient overlays that make work easier. ua-cam.com/video/GKjifMhcvOk/v-deo.html
As a nyer i can verify sd 2.0 was most accurate, we always have headless and shadow people walking around. all our cars are so low to the ground. midjourney wins. its almost unfair to use 2.0 over 1.5, f222 would be even better
Mid journey IMHO has best overall; Dall-e2 for second; and The one I use the most stable diffusion. I prefer stable diffusion 1.4 or 1.5, have not used 2.0+ often enough 10:08 What is up with The cat with hat pins stuck into the side of his face?
i see all these comments about midjouirney. At v4 launch...midjourmey had some issues with the upscale, but the quality of content it was coming up with was amazing. After they did an update, everything looks oversaturated, over polished...it's so not refined. Take side by side comparisons of v4 launch and v4 update...you people are smoking crack. It's not even compareable. Once you notice it, you cant un-notice it. I can look at images on the ranking system and pick out the one's made by the new v4 99% of the time. For illustrations you cant really tell the difference. Im so bummed.... I cant get back any of the variations I made pre-update. Fucking nerfed. I really hope SD catches up...it looks promising. I cant stand all the virtue signalling and liberal authoritarians over at midjourney..it makes me ill.
You can specify you want to use the original V4 vs the update. If you check the announcements channel it showed what flag to use. I thing the sub versions are A and B or something similar.
@@ShawnFumo --style v4a is not the original...its not even close. V4 was comnpletly rewritten. Believe me..I have tried. Go on the community feed and find something you think is really cool created before november 10th. Copy the prompts..make variations...You wont be able to come remotely close to recreating it. I read the feedback section everyday. People are starting to notice the same features...especially with female faces. Thin nose, pointy chin, furrowed brows...it's f'ing terrible. But I appreciatey you trying to help!
DALL-E consistently behind or at the very best barely catching up. MidJourney seems to make everything look amazing, regardless of how much it understands of the prompt. SD somewhere in between, but gives MJ a run for its money on several prompts, especially in picking up clues MJ missed. I don't know how well MJ is configured out of the box, but this doesn't seem like a very good comparison for actual users (aside from beginners maybe). SD requires you to tweak settings and to use things like negative prompts to get the best results. Not as user-friendly, but a lot more powerful.
my vote: photography:MJ4 architecture:SD2 art:D2 design: MJ4 in terms of image quality SD was the most variable one, MJ was the most consistent one, D2 had some better images than the others. SD looks like it would need more help getting a usable image, needs better detailed prompts and more post processing. IF they were all priced the same, for me the winner overall is MJ4.
MJ is pretty much always the best, but the bloom it adds to the shots is distracting. There's a few Dall-E and SD I liked more, but overall MJ takes it.
Haha, you had me blushing! It's not every day I get praised for making people smile and dance at the same time. I'm like a one-man party 🎉🎵, and another vote for Mid-Journey!
Midjourney has the most complete image consistency, but looks overfitted. DALL-E2 seems to have the most original images. Photo: Midjourney Architecture: DALL-E2 Art: dead-heat Design: Midjourney
Though do keep in mind a lot of these prompts are kind of old school with artstation, 8k, and all that. There is also a chaos flag to force the variations in the grid to be more diff from each other. People mentioned SD2 needing negative prompts. As always, each of the AIs have their own techniques and learning curves, even between versions.
hey Han, I'm sorry that you found the video annoying. It's important to keep in mind that everyone has different preferences when it comes to how information is presented, and what one person finds engaging or entertaining may not be the same for someone else. If you found the video frustrating because it didn't get to the point quickly enough or didn't provide the information you were looking for, it might be more helpful to look for other videos or sources of information that better meet your needs. In the future, you can also try using keywords and phrases to search for videos that are more closely aligned with your interests (answered with ChatGPT)
I haven’t used Midjourney since it first came out. *Apparently that was a big mistake.* V4’s performance in this video is *unreal* 😩 What was the price on that again? 🤤
Midjourney took the cake on all categories for me, except for vector logos.... Excellent video, very fun to watch great job!
Thank you very much for participating and voting,
adding up the following votes: 3 for midjourney , 1 for dall-e
How could you not love PUANT POOS!? LOL
I think the same AHAHHA good !
@AlexKongMX same here ✋
damn midjourney crushes on most prompts
Thank you very much for participating and voting,
adding up the following votes: 4 for midjourney
Stable Diffusion 2.0 cant really be compared without using negative prompts.
Is SD2 the only one supporting negative prompts? Looks like it needed some on the chair.
MidJourney does have negative prompts using “-no” (V4 didn’t at first but does now). It also has weighted prompts, image prompts, “remix” (changing the prompt as you choose variations). It does not have in/out painting though.
Stable Diffusion is a do-it-all locally with lots of plug-ins and API's... it's like comparing a 99 in one Swiss army knife to 2 regular knifes of same size and saying it's about the same. oh, and the multi-purpose one is totally free and don't depend on corporate agenda and connectivity :)
i dont get why people praise so much about midjourney, ah yes cuz most of them are noobs
I love freedom software, and free software. But for people who _aren't_ broke, maybe… just maybe… they would rather spend a little money than spend a little time convincing SD not to generate nightmare fuel? Y'know, like it did all over this competition?
Yeah unfortunately you’d have to use the prompts that best fit’s each AI. This is like comparing program languages using the syntax of one and hoping for good results for the others. You must use negatives in SD2 = whole new ball game. As far as more intuitive for the general public then I’d have to give it to MJ. But to literally get amazing works in any of them you’re going to have to use the language prompts the fit with that particular AI.
well said
SD2 is extremely reliant on negative prompts to yield good results. Without them it's generally worse than 1.5.
As long as you specifically ask it to not make anything ugly it's actually really good.
SD2 is worse than SD1.5 atm sadly.
I’d be interested to see how this works out when using stable diffusion with negative prompts since this is the correct way to use it. Also, a test with F222 included for comparison would be good
What about double negative prompts?
photography: midjourney
architecture: stable diffusion
art: midjourney
design: dall-e
I tried a bunch of ai for art and so far midjourney is simply on another level, never been this amazed and lowkey horrified at how fast a computer can give u something so accurate, detailed and stunning.
so one more for MJ
mid-journey V4 is in another level
hey Ahmed: so +1 for MJ
Midjourney understood that the park was a design rather than a photo and gave you full layout. That’s some detail
I've been using Midjourney for a few months now. This just confirms my bias. Clearly the best AI.
so one more for MJ
SD 2.0 is only good with negative prompts.
Midjourney won everything, except Architecture on which SD won.
Thank you very much for participating and voting,
adding up the following votes: 3 for midjourney , 1 for SD
This was such a great comparison video. Thank you!
Midjourney was on point in almost all categories, stable diffusion also gave some interesting results especially in the design category. Dall-e seams to be already behind the competition. Almost never made something better than the other two.
Except for Design logo poster, Midjourney 4 beats the heck out of others. Period. No comparison with MJ4
Except for Design logo poster, Midjourney 4 beats the heck out of others. Period. No comparison with MJ4
You should add dreamlikeart v 1
That one is like midjourney but free...
Stable Diffusion is far from the best, often the worst. But you can run it locally, lots of add ons and custom models goes a long way.
Made me crave for some of those HO-DOGS!! :D
I prefer the relish at Hot Dog and logh, personally.
Midjourney would be king if someday they fix the damn hands. Stable was fated to be the best before it was emasculated.
MJ V5 is aiming to be out by Jan and they said hands will be improved. V4 still has issues but definitely much better than it used to be. Often the mistakes are more subtle now.
@@ShawnFumo Well they just need to get the hands like 50% of times fine and nobody will be able to stand against them.
Midjourney appears to have the best results in this video but as others have pointed out, these are not at all what properly prompted Stable Diffusion 2 images look like. I've used all 3 Ai shown here and this is heavily biased because of the prompts. I doubt that the producer was intentionally doing that, and is probably just a lack of knowledge. That said, Midjourney is indeed great... if you don't mind having to pay for every image you produce, which is something that should also be mentioned. Stable Diffusion is completely free if you have the hardware to run it locally. There is no free option for Midjourney.
Hey Oddar, thanks for watching and commenting on the video! The idea behind this video was to have some fun and see how different models would react to the same prompts. These prompts are not too simple or customized for each model - in fact, they are typically used in stable diffusion. As it turns out, midjourney responds well to simple prompts, as shown in the video. We thought about making a video to show which prompts work best for each model (such as a negative prompt for SD2, for example), but there are probably already plenty of videos out there that explain all that stuff much better than we could. Thank you again for watching and commenting, and I hope you are subscribed and have the notification bell turned on
@@nobodyandthecomputer I did enjoy your video. I was just pointing out that without the correct negative prompts for SD 2 it can suffer quite visibly in image quality. This is far more important for SD 2 than it was for SD 1. People who aren't aware of this will judge those images poorly, when in fact with a generic negative prompt the improvement can be extreme. I quite like Midjourney and think it may indeed have an edge over Stable Diffusion when comparing the base models, but that changes when you use alternate SD models which is something that Midjourney can't do... although that is definitely outside the scope for a generalized video like yours,
I was expecting MJ to win hands down but was really impressed with Dall-e's output. I'm wondering if a lot of people are having too much personal bias towards MJ to be more objective. Given that SD is generally bleh compared to them, a few times it came out impressive.
That’s fair. At least with these prompts, DE2 tends to be more photo real, while MJ injects more artistic sensibility
@@ShawnFumo Yes!
Great video and comparison. Midjourney v4 was all round the best for me. Think they all have fantastic quality though. Exciting times.
Hold onto your popcorn, folks! MJ's about to score again!
Stable diffusion 2 sadly suffers from need to use negative prompts to work, the backlash made the guys from SD to release SD 2.1 which works better without using as much negative prompts.
Don't know if you used amply negative prompts in this test.
No doubt MJ v4 is extremely potent, the best overall for now, nevertheless.
i feel you kinda forgot disco diffusion, but then again disco is more an img2 img tool imo as coherency from it alone suffers alot
As fun as this is, it's not a fair comparison in the least. I love me some mid journey but she is limited on features. SDXL has a ton of extra doodads that you can refine an image with to make something absolutely perfect. Going with the first image isn't really showing the strengths or limitations of each Platform
Midjourney is the best in almost all of the categories. Great video!
And the crowd goes wild, 'cuz MJ's got another one in the bag!
SD version 2.0. OOF. The version they had to patch with 2.1 right after release because it wasn't working.
Ok where did you get the music for this video? Some of these hooks are incredible, the afro-Ghibli one particularly.
it's sad that only stable diffusion can be run natively on a personal computer, but luckily you can actually train stable diffusion to produce midjourney style images lol
If you wouldve used a real voice instead of a robot this couldve been such a good video
it should be called unstable diffusion, artifacts on every pic
sd 1.5 is better then 2.0, competition was unfair
14:56 none, sir. None. Thank you for my 1st ai nightmare.
The forbidden combination of words "Trending on deviantart"
im sipping on the stable diffusers tears in the comments lmao
In truth, SD has no competition for several reasons - first, it can be used locally, it is supported by a community that trains models for it itself, and it has convenient overlays that make work easier. ua-cam.com/video/GKjifMhcvOk/v-deo.html
EY , LOVE THIS EDITION.. GOOD JOB!
thanks Amin!
Midjourney Kicked the fat bottomed.
I thought Midjourney won overall. But the others had better art pieces at times. Awesome video!
excellent Ollie, adding one more for MJ 🫶🏽
Wow, some of music backgrounds are so good
9:52 Wait, do I see correctly that MJ4 generates author signatures and watermarks?
Yes sometimes, though usually they aren’t actually real signatures but just what it thinks a signature for this kind of art would look like.
I love this channel! Respect!
As a nyer i can verify sd 2.0 was most accurate, we always have headless and shadow people walking around. all our cars are so low to the ground. midjourney wins. its almost unfair to use 2.0 over 1.5, f222 would be even better
Mid Journey wins most categories but Dale had some that are clearly better. Photography prompt 4 for example.
so 2 for MJ, 1 for DL2
Mid journey is the king
SD. Imagine paying lol
Mid journey IMHO has best overall; Dall-e2 for second; and The one I use the most stable diffusion. I prefer stable diffusion 1.4 or 1.5, have not used 2.0+ often enough
10:08 What is up with The cat with hat pins stuck into the side of his face?
these tracks were dope
midjourney is OP
i see all these comments about midjouirney. At v4 launch...midjourmey had some issues with the upscale, but the quality of content it was coming up with was amazing. After they did an update, everything looks oversaturated, over polished...it's so not refined. Take side by side comparisons of v4 launch and v4 update...you people are smoking crack. It's not even compareable. Once you notice it, you cant un-notice it. I can look at images on the ranking system and pick out the one's made by the new v4 99% of the time. For illustrations you cant really tell the difference. Im so bummed.... I cant get back any of the variations I made pre-update. Fucking nerfed. I really hope SD catches up...it looks promising. I cant stand all the virtue signalling and liberal authoritarians over at midjourney..it makes me ill.
You can specify you want to use the original V4 vs the update. If you check the announcements channel it showed what flag to use. I thing the sub versions are A and B or something similar.
@@ShawnFumo --style v4a is not the original...its not even close. V4 was comnpletly rewritten. Believe me..I have tried. Go on the community feed and find something you think is really cool created before november 10th. Copy the prompts..make variations...You wont be able to come remotely close to recreating it. I read the feedback section everyday. People are starting to notice the same features...especially with female faces. Thin nose, pointy chin, furrowed brows...it's f'ing terrible. But I appreciatey you trying to help!
Did you train your AI voice on kizzume from jcs?!
Tribvtes from germany
Midjourney
Dall e2
They all do text like it's the far future and every language has merged 😂
Dall e is definetly is the best for creativity
Yippee ki yay! Another vote for DL2!
DALL-E consistently behind or at the very best barely catching up. MidJourney seems to make everything look amazing, regardless of how much it understands of the prompt. SD somewhere in between, but gives MJ a run for its money on several prompts, especially in picking up clues MJ missed.
I don't know how well MJ is configured out of the box, but this doesn't seem like a very good comparison for actual users (aside from beginners maybe). SD requires you to tweak settings and to use things like negative prompts to get the best results. Not as user-friendly, but a lot more powerful.
hey Ramus, so +2 for MJ, +1 for SD
1st category
Midjourney won
2nd category
Dalle beats everyon
3rd category
Midjourney
4th category
Stable diffusion
Thank you very much for participating and voting,
adding up the following votes: 2 for midjourney , 1 for dall-e, 1 for SD
my vote:
photography:MJ4
architecture:SD2
art:D2
design: MJ4
in terms of image quality SD was the most variable one, MJ was the most consistent one, D2 had some better images than the others. SD looks like it would need more help getting a usable image, needs better detailed prompts and more post processing. IF they were all priced the same, for me the winner overall is MJ4.
excellent, adding 2 for MJ, 1 for SD and 1 for D2
MJ is pretty much always the best, but the bloom it adds to the shots is distracting. There's a few Dall-E and SD I liked more, but overall MJ takes it.
Hurray! Another vote for Mid-Journey! It's like every time we get a vote, a unicorn gets its wings 🦄🎉
Midjourney does a very high quality job, Dall-E is not bad, Stable diffusion seems to have the worst quality, except for rare good ideas.
2 votes for Mid-Journey and 1 for DL2!
I am using Midjourney and this is awesome!
Awesome video! Love the bg music and the cuts had me smiling. Good job. 👍
MJ sort of destroyed them in this video. So yes, MJ. ✋
Haha, you had me blushing! It's not every day I get praised for making people smile and dance at the same time. I'm like a one-man party 🎉🎵, and another vote for Mid-Journey!
MidJourney v4 is king!
awesome Virgil!, so adding +1 to MJ
Midjourney won by yards.
thanks Ragip, so one more for MJ!
MJ did get the best results in all prompts - but it got the most "best result" in all categories.
Full points for MJ.
Good effort by others.
fantastic, one more vote for MJ
Stable Dissusion for me
great Duke, +1 for SD
I remember a time when Dall-e 2 was the best hahaha
MidJourney did a great job in almost every category. My vote goes for MJ.
Great video btw.
great, then adding a vote for MJ
2 for mid journey here
adding two more King
Mid journey on all of them, especially the human subjects or faces in general captures emotion better than all the other models I believe
great creatures! so one more vote for MJ😎
In my opinion Midjourney is clearly superior to the competition at this point in time!
so one more for MJ
Midjourney has the most complete image consistency, but looks overfitted. DALL-E2 seems to have the most original images.
Photo: Midjourney
Architecture: DALL-E2
Art: dead-heat
Design: Midjourney
cool, so 2 for MJ, 1 for D2
Though do keep in mind a lot of these prompts are kind of old school with artstation, 8k, and all that. There is also a chaos flag to force the variations in the grid to be more diff from each other. People mentioned SD2 needing negative prompts. As always, each of the AIs have their own techniques and learning curves, even between versions.
Midjourney won
so one more for MJ
midjourney wins
cool, adding one more for MJ
NICE TUTORIAL FOR BEGINNERS
midjourney
great Ajs, +1 for MJ
Mid Journey .. kills it.
1. Midjourney v4
2. dall e 2
3. stable diffusion 2.0
great Mir!, so 2 more for MJ and 1 for DL2 👍
MidJourney V4 hands down!
so one more for MJ!,
great content man, waiting the next video
Thank ignorante , we are working on two scripts for the next videos, they will be great, you'll see
Midjourney it is!
... and one more for MJ, ☝
What an annoying way to tell information. Get to the point... All this nonsense vidoes without getting the point across.
hey Han, I'm sorry that you found the video annoying. It's important to keep in mind that everyone has different preferences when it comes to how information is presented, and what one person finds engaging or entertaining may not be the same for someone else. If you found the video frustrating because it didn't get to the point quickly enough or didn't provide the information you were looking for, it might be more helpful to look for other videos or sources of information that better meet your needs. In the future, you can also try using keywords and phrases to search for videos that are more closely aligned with your interests (answered with ChatGPT)
I haven’t used Midjourney since it first came out. *Apparently that was a big mistake.* V4’s performance in this video is *unreal* 😩
What was the price on that again? 🤤
I think the minimum on Midjourney is $10 a month. 😉
Can't wait a few years from now when when it becomes more accurate to the point it can get the fingers right
MidJourney V5 is aiming to be out probably by Jan and supposed to be better at hands.