Tip. Works with all diffusion art. If you wanna test tokens/ details adherence. Go down the body as you add. Work your way to the feet. (Ie. After green sweater…. Red leather belt…. Blue jean shorts … pink socks …. Yellow sneakers…. Etc)
@@FutureTechPilot welcome! I’ll have to try in midjourney but I did a test in sd 1.5 vs SDXL (which I’m guessing midjourney using as well just highly customized and evolving) Sd15 tops out at like 16 total correlated tokens … SDXL can handle about 30.
Nolan, great video. Hope you get well soon! Something I have noticed, to add: 1) The same person appears when you run --style raw and vary --s in 5.2 as well. 2) They appear for a while, then at some --s value another person appears and also stays for a range of values, etc. 3) The same is true for varying --c value!
I hope you recover from your illness! I just discovered your channel just now, it seems thoughtful and friendly so I subscribed. I sympathize with your "Superman vs Batman" problem, MJ seems to still struggle with having two or more characters of the same genre. I spent a day to get a superhero romance book cover of the hero carrying the heroine, refining and refining to get MJ to not bleed over colors or clothing articles from one to the other. Anyway, thanks for talking about this interesting topic, and good luck!
Thanks for this! Always interesting to see progress in MJ. Don't think I'll be re-subscribing to it because I have no compelling need but glad they are evolving. Hope your personal challenges work out ok.❤
After a day trying, I can't say I am enthused. It doesn't look much better than V5, it lost the understanding of many styles, and if you are not really short, faces very quickly began to look distorted. I tried out a good number of things. But anything but closeup photorealistic portraits didn't convince me.
Very cool, thanks. I love how excited you are, so many people in the chats like to complain and expect perfection right this minute. I don’t do midjourney for a living so it’s just fun. Chaos is my favorite thing - it’s crazy. I can’t stop.
I dunno, doesn't look that different from V5. Apparently it has improved prompt understanding and text capabilities but some of the generations I've seen don't really match the prompt much better either. Maybe I just had my expectations high since they said this new version would be comparable to the jump from V3-V4 instead of V4-V5, but right now I'm not even sure if it even matches the latter which is kind of disappointing.
I agree but I think part of it is we all have to re-learn how to prompt. It seems like you have to put in more work to get something interesting. After my first generation in v6, my reaction was just "oh..." The results are a bit more grainy than v5
@@DreamscapeASMRofficial Yeah but I've yet to see anything that really screams "This is V6!" to me yet and I would've thought that we'd get that by now in at least some pictures. Even the rating pictures in V5 had proper hands and enhanced photorealism and you could tell that it'd be a noticeable step up.
If I don't see much difference in my horror images, one thing I noticed right away in my portraits are that the skin is much more natural, even for models, we don't really have that photoshopped look anymore. That's really good !
Always look forward to your videos. I listened to office hours last week, and they surprised me by seeming to not be that concerned about whether we get exactly what we describe or not. Can't figure out why they aren't working on that. Maybe they should concentrate on that next time. Thank you so very much for keeping us up-to-date. I depend on your channel so much. Hope you are soon feeling better and 100 %.😊
💯%. The arrogance of the leadership of MJ is breathtaking. It can't end well for a subscriber-based platform with mega-competition bearing down on them. Listen to what your CUSTOMERS want, MJ.
Try to make something like this :- A man holding laptop and giving a pose for a photoshoot See what you get And see if you can use it for a social media poster Can you!!!!
Thanks for the update!Looking forward to trying out 6.0 but I already see it still has problems with hands. I feel like I spend half my fast hours region varying fingers.
I do not suppose there is any way to use the seed parameter to keep an original scene e.g "woman sitting on a bench" and then adding to that e.g. wearing a red hat. So we replicate the exact first image and then just have the red hat in addition. I tried to do this by prompting the original image but even at maximum image weight it does not get to what I want. Is there any reason in theory why weights could not be much higher?
I'm not a big fan of seeds, and right now there is no way to do what you're looking for. Maybe in the future once they develop a tool for consistent characters. You could try adding a red hat to the image in something like photoshop and then image prompting that into Midjourney
For the superman thing, I tried this prompt and it gave me some decent results : while we look over the shoulder of superman, we see a defeated batman on the ground --ar 16:9 --s 50 --v 6.0 --style raw while we look over the shoulder of superman, we see a defeated batman on the ground --ar 16:9 --s 50 --v 6.0 --style raw
The coherence looks impressive. I'd love a button that allows you to easily re-run prompts made in previous versions, in V6. I tried copy/pasting my last prompt from V5.2 and got an error message... "Invalid parameter. A maximum of 7 multi-prompts is currently supported." Do you know if MJ have included a list of restrictions/limitations? Or is this another great update without a user manual and a personally signed message from King David that reads... "Here's the keys to your new spaceship. "Just watch a youtube video to learn how to drive it."
haha you know what, I had no idea there was a limit on multi-prompts! That's really cool, but also frustrating for sure that there aren't a clear set of limitations. I've definitely never seen that mentioned before so I doubt it's documented anywhere
@@FutureTechPilot Yeah, I've had it a few times now and a few other people have mentioned it in YT comments. I was hoping to convert a lot of previous v5.2 images over to V6 by adding --v 6 to the Vary functions. TBH the results are mostly terrible. V6 doesn't seem to mix well with other versions. Even my fresh V6 prompts are throwing lots of artefacts and a kind of over-sharpening/contrast scratchy effect. Upscaling helps, but even the subtle can change the entire image substantially, esp faces. I've decided to leave V6 until they've added back all the functionality of 5.2. Giving Leonardo AI a run now. Seems way more reliable, stable and functional.
@@morpheus2573The multiprompt limit was throwing me off, I keep having to cut and combine things to get around it. Since V6 is still in alpha I'm hoping they lift that restriction by the time the rest of it gets fully released
Nolan, I am currently working will DALL-E . Needing more consistency of style and images and including these consistent characters within the context of a created background. Seriously considering MIDJOURNEY. Will I be able to achieve my goal in MidJourney?
No, unfortunately. Some people might disagree - you can check out my video on Character Consistency ua-cam.com/video/zgeI_ffCudY/v-deo.html - maybe what you're looking for is technically 'possible' but in my opinion, it would not work without the assistance of Photoshop
on your chrono trigger warrior, I kind of played around with the variations on remix, and started adding in live action, photorealistic and I really got some awesome results, in how higher you put your stylize, how more focused you get on the character, and the lower on stylize (also allways on raw for this) the more it made the character blend in with the surroundings, the best result I got was when I added in dynamic fighting stance in S 250 , one the images really blew my mind
Can you make a video on how to extract or get a design that you generated in flat lay? For example I was playing with different commands and a flat lay was generated, inside that flat lay is an illustration that I'd really love but I don't know how to extract that illustration out of flat lay. The flat lay is at an angle like 45 degrees. So it's not like zoom in and it's solved. Thanks
I think I know what you mean, but I don't think there is a magic solution for you. Maybe watch some Photoshop tutorials - I don't think there is a Midjourney trick that will help you
9:39 I guess Midjourney couldn’t make sense of a tiger wearing purple sunglasses and an orange hat, so it decided to apply these details to the man instead, or to not apply them at all.
You got my curiosity with the explanation (awesome job, by the way...), but my guy, you got my undivided attention when you mentioned "Chrono Trigger!"
@@FutureTechPilot Man, CT is my #1 RPG of all time, FF7 is #2, Xenogears at #3, Parasite Eve (1) at #4...so that should tell you something! I'm a Squaresoft loyalist! Hahaha!
Looks "broken" to me for now, I mean after testing many simple prompts, all the generated images look less creative, less artistic than version 5. It's like if the model training has focused a lot on stock photos content. Or weird param doesn't work like before. Going back to 5.2 for now!
I write down my prompt in chatgpt, ask it to make it detailed and complex, and then ask it to add its own touches to make the image look better (usually an improvement). Copy and paste in Midjourney. Done amazing results. 😂
My initial impression is that where it hasn't been refined to the level of being able to consistently put out nice stuff like v5... Image based prompts might produce some interesting results with v6.
Quick question, I have created an image I like but its in the wrong aspect ratio. How can I replicate the same image but in a different ratio, when I repeat the prompt with the correct ratio the image changes so much. Thanks
Another great insights video. I have found interesting outcomes with negative weightings eg geisha dancing:: red::-.9 (then tested for the range of values up to and including 1
@@adarwinterdror7245 :: is used to multi-prompt. You can then weight the different sections. This includes negative weightings to exclude or reduce the impact of something in an image. For example geisha images are often red or include a fan.
Have you tried art genres or specific artists such as " in the style of". Does not seem to work in V6 (at least as of yesterday) whilst doing a great job in earlier versions
low stylize and style standard (not raw) seem to give marginally better results. TBH, 5.2 seems to have slipped a bit as well. I wonder if there is anything else in play
I am a filmmaker and watching this it feels like the language of V6 is very close to screenwriting. You write a character and a scene. Only the relevant details to the reader. Than you let Midjourney slot machine do it's magic.
You just choose the most appropriate image from the "slot machine", and you just Re-Roll that image or use the variation button. I remember when version 5.2 they said that we must Re-roll our prompts more than 2 times to get almost exactly what we want. Also I think we should stop using Parameters such as raw, stylize, chaos, etc.. They alternate the image. The simpler the prompt the more faithful the outcome.
So here's the problem I have with all versions since 5.0. The "improvement" is subjective. In 5.1 I was making some really interesting interior architecture that versions since can't replicate in creativity. So my feeling is that the versions aren't significantly better, just different. I'd like to see the ai trained on architecture ordering systems and structure, so it knows how to make an actual staircase or gothic vault. Right now it approximates those entities and sort of, kind of, quasi understands architectural systems.
Have you tried making 5.1 architecture gens, and then using those as image prompts in V6? You can even add weight and multiprompting with images now, so that might work if V6 can't generate what you want right out the gate
MJ is definitely getting better at NLP and that will make prompt design more 'natural' and easy (I guess?) as it should be if their intentions is massive adoption. V6 needs to keep expanding on more useful commands instead of limiting them (Ex. /tune for V6), upscaling is getting behind and yeah, surprised but not impressed so far! Hope it gets better, I'm sticking to 5.2, I'm avoiding hyperrealism (not my thing), I mean, reality is awful enough already😅
First impressions: coherence substantially better, but still has trouble with some scenarios. Text now substantially better, though still duffs it about 50% of the time (but now within the ball park of Dall-E on this). More than three words creates problems. Image quality: no grand leap over V 5 so far as I can see yet anyway. My usual subjects look about the same under 6.
My personal opinion on the new version: + beautiful and rich in detail be it in raw mode or normal mode + more realistic, much more realistic + The interaction between background and motif has become significantly better +promting has become better and more sensitive. + 2.7 times faster to create for the same price. -increased problems with hands, eyes and legs - a lot of words that were banned (same prompts that no longer work) - Texture of the clothes is unrealistically dry in combination with water (rain motif). this is just a small example. It's only the alpha version of v6 and optimizations are taking place. I'm excited to see how it develops further. ❤🎉
7:54 I’ll be more impressed if Midjourney could maintain character consistency (according to the character image reference we provide it with), while accurately following your prompts. The characters it produces all look different.
@tripplefives1402 That’s not one bit sustainable if the same character is repeating over and over in different scenes. Also, I can’t just repeat the exact same face in every scene, since the angle and facial expression would need to change in every scene.
It's degraded in my opinion. I tried a series of prompts for images i had made using V5.x , and almost all of them got worse. Moreover, those prompts that I had used for making webpage and ui actually do not produce any ui or webpage in V6
I have just wondered why they waited so long to release this model - I see some improvement, but there are still issues with faces and hands @@FutureTechPilot
There’s a lot to like in V6, but it’s clearly still untweaked. It frequently oversharpens, still makes horrible faces with multiple subjects (unless the prompt demands otherwise), has some scale issues with subjects (test: a closeup of a duck standing next to a cow), and like you experienced in this video, can slip in and out of photorealism, even when prompted for photos. That said, I've been impressed with the improved prompt understanding and its text ability. Placement of subjects is pretty good, and consistent clothing is great, even with two subjects. As a graphic designer, the text support is great to see, though unfortunately Midjourney so far ignores typeface requests (even for the classics), and has a small range of text aesthetics that can repeat. Application of text will also take some getting used to; if the context of the text is not made clear, Midjourney will try to weave it into the image in unexpected ways. For example, with a bird as the subject, it sometimes made the letters out of twine as if it were nesting material. But with packaging,, the use of text is more reliable. Testing a beer can as the subject, the placement of the text was as expected, and it consistently well-balanced, and created a good composition of peripheral elements around the text. Prompting "subtitle" can also help when you want flat text on top of an image (effectively detached from the image). Perhaps you would consider exploring the use of text in a video.
There's a HUGE improvement in textures, lighting, and different styles, though. The whole novelty of 'Wow, it looks so real' across all AI image generators was wearing off for me, but yeah, unless really specified, it doesn't default to photorealism. Retired Creadtive Director here - control over text and typefaces will have to remain in our hands for now. And hands have improved, but not much, YET. Enjoy! :) 👍
I guess you haven’t done any testing with the style tuner yet? What I’m wondering is if I will still be able to use my existing style code or will v6 throw it off even a little causing me to have to completely redo it.
Hi Logan! I always check out your tutorials! V6 looks promising! "The MidJourney slot machine:" so accurate 😆 The mysterious prompting: it's like oracle communication with the gods. You need to be poetic (or high on something). Finally: English is (obviously) not my native language... what does "unsplash" mean? I couldn't have guessed that in a million years 😆 I'm sure you'll recover completely! Best wishes and thanks for the invaluable tutorials! 🤜🏻🤛🏻
Unsplash is a stock images resource (not part of MJ). But most of the time it has a really pleasing effect on the output (at least in v5.2). This is especially true with photorealistic images. Try it out. That's the best way to find out. :)
Mine are displaying small scale nudity based on simple prompts, started with fallen angel and then some other simplistic ones give similar details. What's just as interesting is most things are made with an art painting style without me asking
Can you tell me how to get access to Midjourney? Aside from paying for it. I'd like to test it first. So far I've just been playing with crAIyon, dalle, leonardo and stable
the fact that it turns "more detail=painting" is not a problem as you can always just controlnet it in SD or put it into leonardo photo real. the fact that it gets hands, cigarettes, etc all correct is an achievement in itself. also im really glad they keep the old models out there to use, i kind of like the way version 5 takes things in ways you never imagined, even if its not totally prompt adherent. the new upscale *creative* seems a lot like leonardos 'refine image', it improved the eyes and the hands in one of my generations in much the same way. i took them both in photoshop and mixed and matched them, they were still similar enough to do that.
Yah I should have mentioned it in this video, sorry! V6 is slightly more expensive at the moment, but it's expected to become cheaper as the model gets optimized
I ran out of fast hours....YESTERDAY!!! 🤬 Thanks for the update. Feel better. Who typed it in Discord, huh? Huh? Who did? 😆 Well, pretty much everyone and their mother figured it would be available this weekend. Now to wait for my fast hours!
I haven´t checked MJ v6 yet. But the reflection in the eye probably didn´t work, because your prompt was too much of how a human would think. First rule of promting: Words lose priority from left to right. So the prompt should start with the main subject. The eye. Next step: reflection and so on and so forth. Don´t forget you can weight single words as well but all word weights together should not exceed the sum of 1. In Dall E 3, it perfectly works with That rule goes for all Text to image AIs. General rules of thumb: Order of words. As described. Amount of words: The more words you use, the less weight every word has. Emphasizing on a word works best by either decreasing all other words via weighting or typing the word multiple times. Too many words will burn the image. You can counter that by decreasing the CFG-scale in stable diffusion or the stylize value in Midjourney. Or decreasing word weights further. Sacrificing less important words might do a lot. Don´t use whole sentences. Words like "and" are effective when telling the AI to really do both and not just a mix of both. But for example "an A inside of a B" takes a lot more weight than a simple "A inside B" or even just "A in B". Don´t forget to use negative prompts as well. Here the same rules apply. When wanting to transport a concept onto another, words like "reimagined as", "in the style of" work better than just adding them together with a "-". Hybrids of things often work best by simply typing both things and start playing with weights. words that have similar reasons or are being used in the same context are connected very close to each other in the latent space. So you can push thing into directions. Example: Prompting for an old grumpy man with a blonde wig, a blue suit and orange facetan will most likely generate exactly Donald trump. Even if you didn´t ask for him in particular. Judging by what you have shown, MJ 6 is similarly capable as Dall E 3, so they handly more words better and even understand more complex concepts. But in case, something doesn´t work right, those rules still help improving results. My general prompt construction goes like this: important concept/feature of/activity of > point of view (is basicly a concept) > main subject > subject properties like color, material, shape > Background, additional details.
I would like to see a head-to-head comparison of MJ6 and DALL•E 3. A Unicorn riding a Unicycle is always a test for me. I still cannot get ANY AI art generator to create a LaMancha Goat - always get ears and horns.
V6 is very similar to Dalle3 in what it generates , try the using same verbose dalle 3 prompts in midjourney ,you'll find that it will yield crazy detailed results .
I've been a great MJ fan but this seems underwhelming. Your video cassette text example was impressive but would love to know how many attempts you made. My efforts with text show no improvement over V5 and this aspect is Waaaaaaaaaaay behind DALL-E. Are you going to do a video on the alpha web option soon. Probably not that many users have generated the requisite 20,000 images but I guess that will ease soon. There are pluses and minuses compared with discord, in my view
They specifically asked people not to make content about the Alpha Website because it's only available to certain people, and they don't want their users rushing generations just to unlock it. So once they open it up to more people, I'll make a video on it for sure
Tip. Works with all diffusion art. If you wanna test tokens/ details adherence. Go down the body as you add. Work your way to the feet. (Ie. After green sweater…. Red leather belt…. Blue jean shorts … pink socks …. Yellow sneakers…. Etc)
Hey that's a cool idea! Thanks
@@FutureTechPilot welcome! I’ll have to try in midjourney but I did a test in sd 1.5 vs SDXL (which I’m guessing midjourney using as well just highly customized and evolving)
Sd15 tops out at like 16 total correlated tokens … SDXL can handle about 30.
Nolan, great video. Hope you get well soon!
Something I have noticed, to add:
1) The same person appears when you run --style raw and vary --s in 5.2 as well.
2) They appear for a while, then at some --s value another person appears and also stays for a range of values, etc.
3) The same is true for varying --c value!
Yah I think you're right! It's hard to know when that change will occur haha
I hope you recover from your illness! I just discovered your channel just now, it seems thoughtful and friendly so I subscribed. I sympathize with your "Superman vs Batman" problem, MJ seems to still struggle with having two or more characters of the same genre. I spent a day to get a superhero romance book cover of the hero carrying the heroine, refining and refining to get MJ to not bleed over colors or clothing articles from one to the other. Anyway, thanks for talking about this interesting topic, and good luck!
Thanks a lot for the kind words and sharing your thoughts!
Praying for your speedy recovery! You gotta get healthy and keep bringing these amazing videos mister:)
Thanks for this! Always interesting to see progress in MJ. Don't think I'll be re-subscribing to it because I have no compelling need but glad they are evolving. Hope your personal challenges work out ok.❤
Thanks!
Thank you!!!
After a day trying, I can't say I am enthused. It doesn't look much better than V5, it lost the understanding of many styles, and if you are not really short, faces very quickly began to look distorted. I tried out a good number of things. But anything but closeup photorealistic portraits didn't convince me.
It's a good thing this is only an alpha!
Great for making action photography and paintings for the moment. Will have to take more time. Great video as always. Hope u doing good now.
Yes doing much better. Thanks for the comment!
Im still testing it and very happy to see ur video with good informations❤ thank u
Very cool, thanks. I love how excited you are, so many people in the chats like to complain and expect perfection right this minute. I don’t do midjourney for a living so it’s just fun. Chaos is my favorite thing - it’s crazy. I can’t stop.
Yes! #TeamChaos !!
I dunno, doesn't look that different from V5. Apparently it has improved prompt understanding and text capabilities but some of the generations I've seen don't really match the prompt much better either. Maybe I just had my expectations high since they said this new version would be comparable to the jump from V3-V4 instead of V4-V5, but right now I'm not even sure if it even matches the latter which is kind of disappointing.
I agree but I think part of it is we all have to re-learn how to prompt. It seems like you have to put in more work to get something interesting. After my first generation in v6, my reaction was just "oh..." The results are a bit more grainy than v5
@@DreamscapeASMRofficial Yeah but I've yet to see anything that really screams "This is V6!" to me yet and I would've thought that we'd get that by now in at least some pictures. Even the rating pictures in V5 had proper hands and enhanced photorealism and you could tell that it'd be a noticeable step up.
Theres a huge diference in realism, but the part that blows me away is composition. Images look much more 3d
I've only become more and more impressed with it since launch
If I don't see much difference in my horror images, one thing I noticed right away in my portraits are that the skin is much more natural, even for models, we don't really have that photoshopped look anymore. That's really good !
Awesome!
Always look forward to your videos. I listened to office hours last week, and they surprised me by seeming to not be that concerned about whether we get exactly what we describe or not. Can't figure out why they aren't working on that. Maybe they should concentrate on that next time. Thank you so very much for keeping us up-to-date. I depend on your channel so much. Hope you are soon feeling better and 100 %.😊
💯%. The arrogance of the leadership of MJ is breathtaking. It can't end well for a subscriber-based platform with mega-competition bearing down on them. Listen to what your CUSTOMERS want, MJ.
Try to make something like this :-
A man holding laptop and giving a pose for a photoshoot
See what you get
And see if you can use it for a social media poster
Can you!!!!
Thanks for the update!Looking forward to trying out 6.0 but I already see it still has problems with hands. I feel like I spend half my fast hours region varying fingers.
I hope it helps - I gave you a thumbs-up.
Thanks for the update - and get well soon x
Your enthusiasm is infectious! Always look forward to your videos. Hope you feel better.
Thanks for the kind words!!
I do not suppose there is any way to use the seed parameter to keep an original scene e.g "woman sitting on a bench" and then adding to that e.g. wearing a red hat. So we replicate the exact first image and then just have the red hat in addition. I tried to do this by prompting the original image but even at maximum image weight it does not get to what I want. Is there any reason in theory why weights could not be much higher?
I'm not a big fan of seeds, and right now there is no way to do what you're looking for. Maybe in the future once they develop a tool for consistent characters.
You could try adding a red hat to the image in something like photoshop and then image prompting that into Midjourney
For the superman thing, I tried this prompt and it gave me some decent results :
while we look over the shoulder of superman, we see a defeated batman on the ground --ar 16:9 --s 50 --v 6.0 --style raw while we look over the shoulder of superman, we see a defeated batman on the ground --ar 16:9 --s 50 --v 6.0 --style raw
Woah, cool idea. Thanks a lot for sharing!
The coherence looks impressive. I'd love a button that allows you to easily re-run prompts made in previous versions, in V6. I tried copy/pasting my last prompt from V5.2 and got an error message...
"Invalid parameter. A maximum of 7 multi-prompts is currently supported." Do you know if MJ have included a list of restrictions/limitations? Or is this another great update without a user manual and a personally signed message from King David that reads... "Here's the keys to your new spaceship. "Just watch a youtube video to learn how to drive it."
haha you know what, I had no idea there was a limit on multi-prompts! That's really cool, but also frustrating for sure that there aren't a clear set of limitations. I've definitely never seen that mentioned before so I doubt it's documented anywhere
@@FutureTechPilot Yeah, I've had it a few times now and a few other people have mentioned it in YT comments. I was hoping to convert a lot of previous v5.2 images over to V6 by adding --v 6 to the Vary functions. TBH the results are mostly terrible. V6 doesn't seem to mix well with other versions. Even my fresh V6 prompts are throwing lots of artefacts and a kind of over-sharpening/contrast scratchy effect. Upscaling helps, but even the subtle can change the entire image substantially, esp faces. I've decided to leave V6 until they've added back all the functionality of 5.2. Giving Leonardo AI a run now. Seems way more reliable, stable and functional.
@@morpheus2573The multiprompt limit was throwing me off, I keep having to cut and combine things to get around it. Since V6 is still in alpha I'm hoping they lift that restriction by the time the rest of it gets fully released
@@Unbridled-Whimsy 💯%
Nolan, I am currently working will DALL-E . Needing more consistency of style and images and including these consistent characters within the context of a created background. Seriously considering MIDJOURNEY. Will I be able to achieve my goal in MidJourney?
No, unfortunately. Some people might disagree - you can check out my video on Character Consistency ua-cam.com/video/zgeI_ffCudY/v-deo.html - maybe what you're looking for is technically 'possible' but in my opinion, it would not work without the assistance of Photoshop
on your chrono trigger warrior, I kind of played around with the variations on remix, and started adding in live action, photorealistic and I really got some awesome results, in how higher you put your stylize, how more focused you get on the character, and the lower on stylize (also allways on raw for this) the more it made the character blend in with the surroundings, the best result I got was when I added in dynamic fighting stance in S 250 , one the images really blew my mind
dang, I'm gonna go try that right now!!! Thanks for the sweet tip
Curious to see What u got
Can you make a video on how to extract or get a design that you generated in flat lay? For example I was playing with different commands and a flat lay was generated, inside that flat lay is an illustration that I'd really love but I don't know how to extract that illustration out of flat lay. The flat lay is at an angle like 45 degrees. So it's not like zoom in and it's solved. Thanks
I think I know what you mean, but I don't think there is a magic solution for you. Maybe watch some Photoshop tutorials - I don't think there is a Midjourney trick that will help you
mj isn't able to create tape (cassette) and vhs as I see, I've tried tons of styles in v5. 2 and tapes had always the wrong geometry
haha that's always been funny to me (how MJ clearly knows what you're referring to, but somehow messes it up in a weird way)
9:39 I guess Midjourney couldn’t make sense of a tiger wearing purple sunglasses and an orange hat, so it decided to apply these details to the man instead, or to not apply them at all.
Yah right around there seemed to be the limit!
You got my curiosity with the explanation (awesome job, by the way...), but my guy, you got my undivided attention when you mentioned "Chrono Trigger!"
hahaha I'm glad other people have a soft spot for that Franchise
@@FutureTechPilot Man, CT is my #1 RPG of all time, FF7 is #2, Xenogears at #3, Parasite Eve (1) at #4...so that should tell you something! I'm a Squaresoft loyalist! Hahaha!
Looks "broken" to me for now, I mean after testing many simple prompts, all the generated images look less creative, less artistic than version 5. It's like if the model training has focused a lot on stock photos content. Or weird param doesn't work like before. Going back to 5.2 for now!
Did you try adding things like creative/artistic/stylised etc or specifying the style you want to the prompts?
@@dmarsub Yeah, and I can guarantee it's night and day on the majority of the generations.
I write down my prompt in chatgpt, ask it to make it detailed and complex, and then ask it to add its own touches to make the image look better (usually an improvement). Copy and paste in Midjourney. Done amazing results. 😂
Version 5 prompts definitely don't work the same in V 6!
My initial impression is that where it hasn't been refined to the level of being able to consistently put out nice stuff like v5... Image based prompts might produce some interesting results with v6.
Yah I think V6 will go through lots of improvements in the next few weeks
I noticed you didn’t use :: or image weight numbers, does v6 generate better images with commas? Thank you for this video!
I'm not sure if it's 'better', but multi-prompting with :: doesn't seem as necessary anymore. Still lots of testing needs to be done!
You can’t zoom out 1.4x on discord . Did the take that feature away. Or maybe it’s because it’s new they will add it later ?
It's just not available in version 6 yet!
Quick question, I have created an image I like but its in the wrong aspect ratio. How can I replicate the same image but in a different ratio, when I repeat the prompt with the correct ratio the image changes so much. Thanks
You could do this in Version 5 with 'custom zoom' but that feature isn't available yet in V6
Another great insights video. I have found interesting outcomes with negative weightings eg geisha dancing:: red::-.9 (then tested for the range of values up to and including 1
Can you explain what are those :: and -9 you wrote? What does it do?
@@adarwinterdror7245 :: is used to multi-prompt. You can then weight the different sections. This includes negative weightings to exclude or reduce the impact of something in an image. For example geisha images are often red or include a fan.
Have you tried art genres or specific artists such as " in the style of". Does not seem to work in V6 (at least as of yesterday) whilst doing a great job in earlier versions
I experimented with that only a little bit, and yeah it wasn't working too well. Let me know if you have found any tips!
low stylize and style standard (not raw) seem to give marginally better results. TBH, 5.2 seems to have slipped a bit as well. I wonder if there is anything else in play
I am a filmmaker and watching this it feels like the language of V6 is very close to screenwriting.
You write a character and a scene. Only the relevant details to the reader.
Than you let Midjourney slot machine do it's magic.
0:38 The text may be better than before, but the kerning is kaos.
“Same image, JUST a different tattoo”, ha ha ha!
You just choose the most appropriate image from the "slot machine", and you just Re-Roll that image or use the variation button. I remember when version 5.2 they said that we must Re-roll our prompts more than 2 times to get almost exactly what we want. Also I think we should stop using Parameters such as raw, stylize, chaos, etc.. They alternate the image. The simpler the prompt the more faithful the outcome.
Yah if you have a very specific prompt, extra parameters are probably not helpful
I've noticed that a lot of models have issues with cinematic camera angles. "over the shoulder" shots in particular.
Yeah I bet it has something to do with how the data is actually labeled
You really got a figure to sit in a car seat , and haveing a head pop out of the hood 😄😄
😂 AI is hilarious
I can't comprehend how good this has gotten in 12 months.
Yah it's truly mindblowing
Thank you for your detailed interpretation, from Hamburg, Germany, Reinhard (Ron)
Cheers!
I’m so excited ❤
Yess, enjoy!
I love your vids, get well soon Nolan 😊
haha thanks for the support!
I just used V6 and wow!!! I am satisfied with it so far!!! Great vid dude!!!
What would u consider the big improvements in v6?
So here's the problem I have with all versions since 5.0. The "improvement" is subjective. In 5.1 I was making some really interesting interior architecture that versions since can't replicate in creativity. So my feeling is that the versions aren't significantly better, just different. I'd like to see the ai trained on architecture ordering systems and structure, so it knows how to make an actual staircase or gothic vault. Right now it approximates those entities and sort of, kind of, quasi understands architectural systems.
Yeah you're right, each model has its own tradeoffs
Have you tried making 5.1 architecture gens, and then using those as image prompts in V6? You can even add weight and multiprompting with images now, so that might work if V6 can't generate what you want right out the gate
MJ is definitely getting better at NLP and that will make prompt design more 'natural' and easy (I guess?) as it should be if their intentions is massive adoption. V6 needs to keep expanding on more useful commands instead of limiting them (Ex. /tune for V6), upscaling is getting behind and yeah, surprised but not impressed so far! Hope it gets better, I'm sticking to 5.2, I'm avoiding hyperrealism (not my thing), I mean, reality is awful enough already😅
First impressions: coherence substantially better, but still has trouble with some scenarios. Text now substantially better, though still duffs it about 50% of the time (but now within the ball park of Dall-E on this). More than three words creates problems. Image quality: no grand leap over V 5 so far as I can see yet anyway. My usual subjects look about the same under 6.
It definitely cannot handle a lot of words right now, but I know they're working on an update that we will see soon!
Thanks mate
Cheers pal!
Unfortunately, there’s no higher resolution in v6, I’ve expected 2048k
Had no problem getting 2048 via the included upscaling
@@Slen_ upscaling is not same, it can’t add more details and fix finger problems
Finger problems will be able to be fixed after beta, anyway; they never said this is the full/final v6. It is a beta.@@Berlioz1991
Yah I know their team isn't happy with that either, but it's something they're actively working on for sure
I didnt know I needed Tom Cruise as a sith. I need that.
😂
Cool video thanks !!!
Cheers!
1:30
And getting g acces to it is even tougher
Can you show us how to create images in a hexagon canvas instead of a rectangle canvas.
That would be pretty cool but you can't create in shapes like that!
My personal opinion on the new version:
+ beautiful and rich in detail be it in raw mode or normal mode
+ more realistic, much more realistic
+ The interaction between background and motif has become significantly better +promting has become better and more sensitive.
+ 2.7 times faster to create for the same price.
-increased problems with hands, eyes and legs
- a lot of words that were banned (same prompts that no longer work)
- Texture of the clothes is unrealistically dry in combination with water (rain motif). this is just a small example.
It's only the alpha version of v6 and optimizations are taking place. I'm excited to see how it develops further. ❤🎉
Thanks for the breakdown! I haven't come across any banned words yet but I've heard it's been frustrating for some people
@@FutureTechPilot yes but i wrote with the support and i can understand the reasons. Thanks for ur answer😉and good videos.
7:54 I’ll be more impressed if Midjourney could maintain character consistency (according to the character image reference we provide it with), while accurately following your prompts. The characters it produces all look different.
@tripplefives1402 That’s not one bit sustainable if the same character is repeating over and over in different scenes. Also, I can’t just repeat the exact same face in every scene, since the angle and facial expression would need to change in every scene.
At least we know they're working on a feature like that!
want to ask, how to zoom out in v6 ??
That feature didn't make it into the launch, but they're working on it
It's degraded in my opinion. I tried a series of prompts for images i had made using V5.x , and almost all of them got worse. Moreover, those prompts that I had used for making webpage and ui actually do not produce any ui or webpage in V6
Yah they said we have to learn to prompt differently, and my brain hasn't quite fully grasped it yet
why I can't put tattoo in face in midjourney?
It can be frustrating for sure. Try watching this video here - ua-cam.com/video/hPvGRKilnV8/v-deo.html for some tips
Waiting for the pan function in V6
hundo, can't wait for all the features to be available
Hi!
Do you have an affiliate link/ discount code for Midjourney by chance?
Nope, they don't do affiliate links - don't want their users to become salespeople
but "Zoom Out" and "Vary (Region)" are gone now 😧
They'll be back!
@@FutureTechPilot awesome! V6 has been blowing my mind!
Just tried it. Thanks for the heads up! I left MJ for Ideogram because of the text issues and now I may come back.
haha Ideogram is probably better at text right now, but Midjourney can still improve!
Thanks for the recap. This looks promising.
no upscaller, low resolution images we can't use. Not in my case anyway....
And it is not compatible with style tuner???? Hope this won’t stay that way!!!!
They're definitely working on a V6 style tuner
I wonder how far one can go by switching between versions when doing remixes.
I've definitely wanted to experiment with that myself!
So 17 parameters? I'm I counting good? 18-19 stretching
I don't think there is actually a hard number - seems to have a lot of variance depending on the ideas in the prompt
I am testing it - it gives more realistic results:)
Oh heck yah it does
I have just wondered why they waited so long to release this model - I see some improvement, but there are still issues with faces and hands @@FutureTechPilot
There’s a lot to like in V6, but it’s clearly still untweaked. It frequently oversharpens, still makes horrible faces with multiple subjects (unless the prompt demands otherwise), has some scale issues with subjects (test: a closeup of a duck standing next to a cow), and like you experienced in this video, can slip in and out of photorealism, even when prompted for photos. That said, I've been impressed with the improved prompt understanding and its text ability. Placement of subjects is pretty good, and consistent clothing is great, even with two subjects. As a graphic designer, the text support is great to see, though unfortunately Midjourney so far ignores typeface requests (even for the classics), and has a small range of text aesthetics that can repeat. Application of text will also take some getting used to; if the context of the text is not made clear, Midjourney will try to weave it into the image in unexpected ways. For example, with a bird as the subject, it sometimes made the letters out of twine as if it were nesting material. But with packaging,, the use of text is more reliable. Testing a beer can as the subject, the placement of the text was as expected, and it consistently well-balanced, and created a good composition of peripheral elements around the text. Prompting "subtitle" can also help when you want flat text on top of an image (effectively detached from the image). Perhaps you would consider exploring the use of text in a video.
There's a HUGE improvement in textures, lighting, and different styles, though. The whole novelty of 'Wow, it looks so real' across all AI image generators was wearing off for me, but yeah, unless really specified, it doesn't default to photorealism. Retired Creadtive Director here - control over text and typefaces will have to remain in our hands for now. And hands have improved, but not much, YET.
Enjoy! :) 👍
Hope this makes you feel better, when I saw it had released I ran to UA-cam to see if you had uploaded something. (:
hahah and that's why I wake up at 2:30 am just to make a video for all of you!
great video
Thanks for the comment!
I guess you haven’t done any testing with the style tuner yet? What I’m wondering is if I will still be able to use my existing style code or will v6 throw it off even a little causing me to have to completely redo it.
I try, but it's hit or miss whether v6 works with my style tuner. Some of them work, some of them cause errors.
@@clfrancis Ugh.
Hi Logan! I always check out your tutorials! V6 looks promising! "The MidJourney slot machine:" so accurate 😆 The mysterious prompting: it's like oracle communication with the gods. You need to be poetic (or high on something). Finally: English is (obviously) not my native language... what does "unsplash" mean? I couldn't have guessed that in a million years 😆 I'm sure you'll recover completely! Best wishes and thanks for the invaluable tutorials! 🤜🏻🤛🏻
Unsplash is a stock images resource (not part of MJ). But most of the time it has a really pleasing effect on the output (at least in v5.2). This is especially true with photorealistic images. Try it out. That's the best way to find out. :)
@@AG_before many thanks!!!
Thanks for posting even though you’re not recovered yet!
❤ Editing this video was so hard lol I was nauseous and felt awful all morning. So I appreciate the kind words
If the way of writing the prompt is different in v6, then you can't compare v5.2 with v6 with the same prompt...don't you think?
right, and I was trying to show what would happen if you did that
Mine are displaying small scale nudity based on simple prompts, started with fallen angel and then some other simplistic ones give similar details. What's just as interesting is most things are made with an art painting style without me asking
I've heard about the accident nudity being more common now. I think that will get patched up with the next update
Can you tell me how to get access to Midjourney? Aside from paying for it. I'd like to test it first. So far I've just been playing with crAIyon, dalle, leonardo and stable
Nope. Have to pay. But you can cancel anytime.
Yeah unfortunately they had to disable free trials because people found an exploit to abuse them
Im assuming all this is still accessed by the discord?
Yeah! but they're working on a website for creation
@FutureTechPilot
Oh nice. They keep everyone in the discord up to date with news?
And how much is it? Are there different tiers?
the fact that it turns "more detail=painting" is not a problem as you can always just controlnet it in SD or put it into leonardo photo real. the fact that it gets hands, cigarettes, etc all correct is an achievement in itself.
also im really glad they keep the old models out there to use, i kind of like the way version 5 takes things in ways you never imagined, even if its not totally prompt adherent.
the new upscale *creative* seems a lot like leonardos 'refine image', it improved the eyes and the hands in one of my generations in much the same way. i took them both in photoshop and mixed and matched them, they were still similar enough to do that.
Yeah always having access to older models is a big selling point for me
Its good..but some result are better in 5.2 ..specially skin tone..
Thank you for the production and uploads, Nolan.😀
And thank you for the comment!
@@FutureTechPilot Merry Christmas and get well soon, Nolan.
Thanks for sharing
What is the cost difference between v6 and v5?
Yah I should have mentioned it in this video, sorry! V6 is slightly more expensive at the moment, but it's expected to become cheaper as the model gets optimized
I ran out of fast hours....YESTERDAY!!! 🤬
Thanks for the update. Feel better. Who typed it in Discord, huh? Huh? Who did? 😆 Well, pretty much everyone and their mother figured it would be available this weekend.
Now to wait for my fast hours!
I haven´t checked MJ v6 yet. But the reflection in the eye probably didn´t work, because your prompt was too much of how a human would think. First rule of promting: Words lose priority from left to right. So the prompt should start with the main subject. The eye. Next step: reflection and so on and so forth. Don´t forget you can weight single words as well but all word weights together should not exceed the sum of 1.
In Dall E 3, it perfectly works with
That rule goes for all Text to image AIs.
General rules of thumb:
Order of words. As described.
Amount of words: The more words you use, the less weight every word has. Emphasizing on a word works best by either decreasing all other words via weighting or typing the word multiple times.
Too many words will burn the image. You can counter that by decreasing the CFG-scale in stable diffusion or the stylize value in Midjourney. Or decreasing word weights further. Sacrificing less important words might do a lot.
Don´t use whole sentences. Words like "and" are effective when telling the AI to really do both and not just a mix of both.
But for example "an A inside of a B" takes a lot more weight than a simple "A inside B" or even just "A in B".
Don´t forget to use negative prompts as well. Here the same rules apply.
When wanting to transport a concept onto another, words like "reimagined as", "in the style of" work better than just adding them together with a "-". Hybrids of things often work best by simply typing both things and start playing with weights.
words that have similar reasons or are being used in the same context are connected very close to each other in the latent space. So you can push thing into directions.
Example: Prompting for an old grumpy man with a blonde wig, a blue suit and orange facetan will most likely generate exactly Donald trump. Even if you didn´t ask for him in particular.
Judging by what you have shown, MJ 6 is similarly capable as Dall E 3, so they handly more words better and even understand more complex concepts. But in case, something doesn´t work right, those rules still help improving results.
My general prompt construction goes like this:
important concept/feature of/activity of > point of view (is basicly a concept) > main subject > subject properties like color, material, shape > Background, additional details.
I would like to see a head-to-head comparison of MJ6 and DALL•E 3. A Unicorn riding a Unicycle is always a test for me. I still cannot get ANY AI art generator to create a LaMancha Goat - always get ears and horns.
my favorite part is when you couldn't think of more colors besides red, green, blue and yellow 🤣
😂 I got stumped
Hi, what about zoom?
didn't make it in at launch, but they're working on it
yesterday night till 2 o clock i was exploring v6 and i am so excited and thrilled, output value is just amazing !
I agree!
Bhai 6000 me gir raha hai..1 month ka. 😢😢
This is just absolutely amazing! Great video dude!
Thanks a lot for the comment!
Amazing v6. Thanx for all the videos ! :)
And thank you for the comment!
i am new in midjourney AI and want to buy subscription , can u please guide how it works and how i will upgrade it to version 6 ,, please guide me
You can check out this video - ua-cam.com/video/X8l7UJYO4u4/v-deo.html - and Version 6 is available to everyone with a subscription
@@FutureTechPilot thank you😊
i really wish they would release a video generator
@@ryanbigguy where have they said that at?
They're definitely working on one
Thank you so much for all the free PDfs, very good indeed, please going. Please keep posting V6 tips. I'm a fan of midjourney.
Thanks a lot for the feedback! I've got lots more video's planned
it doesn't work so far
Lmao my man got a easy new sub by subtlety dropping "The Interview" quote "same same, but different" BAHAHA
😂 😂 that's the kind of culture I want
V6 is very similar to Dalle3 in what it generates , try the using same verbose dalle 3 prompts in midjourney ,you'll find that it will yield crazy detailed results .
Idk, you can’t even generate a Batman with dalle, the content policy is so restrictive it’s almost useless
Yeah the style of prompting works really well
I don't see too much of a difference. The new DALL-E still got it for me.
The difference is effing _massive_
After spending a lot of time with it, I am continually blown away by what it can do lol
I've been a great MJ fan but this seems underwhelming. Your video cassette text example was impressive but would love to know how many attempts you made. My efforts with text show no improvement over V5 and this aspect is Waaaaaaaaaaay behind DALL-E.
Are you going to do a video on the alpha web option soon. Probably not that many users have generated the requisite 20,000 images but I guess that will ease soon. There are pluses and minuses compared with discord, in my view
They specifically asked people not to make content about the Alpha Website because it's only available to certain people, and they don't want their users rushing generations just to unlock it. So once they open it up to more people, I'll make a video on it for sure
Fair enough. Looking forward to it. BTW you do a great job replying to messages
Thank you once again! ❤
Cheers!
v 6 IS SOOOO MUCH better than before
The first text didn't turn out perfectly. Notice the letter spacings, they vary quite a much
ya maybe!