BIG Update - Midjourney Will Never Be the Same Again
ฝัง
- เผยแพร่เมื่อ 29 มิ.ย. 2024
- Check out Midjourney's NEWEST feature - Personalization - I'll show you how to train the bot and different ways you can implement your own style in Version 6 of Midjourney!
Use '--p woml1ai' in your prompt for my 'style'!
Midjourney Beginner Course ➡️ ftp.live/mjcourse
Podcast Link ➡️ / @futuretechpodcast
► FREE PDFS
Prompt Glossary V6 ➡️ ftp.live/V6Glossary
2-4-2 V6 ➡️ ftp.live/242
200 Aesthetics ➡️ ftp.live/200Styles
Midjourney Cheat Sheet ➡️ ftp.live/CheatSheet
► SOCIAL
Discord - / discord
Instagram: / futuretechpilot
Timestamps:
0:00 Intro
0:20 Example 1
0:58 How to Trigger Personalization
1:14 Example 2
1:48 How to Train the Bot
3:29 How to Test Your Style
4:52 How to Control the Strength
7:52 Experiment with Chaos
9:41 Aspect Ratios
10:22 Short vs. Long Prompts
11:18 Style Raw
Midjourney really needs a feature for posing multiple consistent characters together. We should be able to combine multiple character references when generating an image.
Exactly
There's another AI discord bot called faceswap for that.
Just better control for image to image would be great already
I would resubscribe instantly if they added something like controlnet with edge to image, part of my workflow is doing 3d renders and "reskinning" them with ai . Without that midjourney is useless for me
@@rastafabui yeah the image to image thing they have is hot garbage. and no depth/pose/edge to image which restricts creativity and possibilities drastically
Awesome video. I just discovered the, personalization --p, earlier today but haven't had time to check it out. Thank you for examining the new features so thoroughly and explaining them so well in all your videos. I never miss a single one of your videos. Excellent work!
What mid journey should release is multiple personalizations.
Sometimes I feel like hyper realism, sometimes I feel like illustration, sometimes I feel like sketchy, sometimes I have a client who needs something else.
they already have that (and this feature shown in the video barely seems like a big update). try this example (you can modify the prompt)
/tune prompt digital illustration with thick lines and bold colors --ar 16:9
Choose 16 Style Directions to get a sense of what this does, and you should be good to go from there. Get as many styles you want using different prompts
@@ExploringThinkerAs far as I know that feature isn’t compatible with V6 and only works with V5.2
I miss the style tuner pretty much!!!! Why didn’t they release it for v6?
Fantastic explanation! Thanks so much for making it fun and easy. :)
Hey I'm glad I could help!
Oh WOW, can't wait to try, but I def have to train the bot for my style a bit more first. Thanks Nolan, as usual!
Thank you so much Nolan for always sharing what's going on with Midjourney! You're the best!
I was so excited when I heard they were working on personalization and this just feels like a hustle sorry to say.
Midjourney let you create your own style codes without needing you to go through hundreds of other people's pictures that have nothing to do with your style. Personalization should be you going through your own pictures and tagging the ones you like specifically for personalization. They should call it "please do free work for us" lol I would respect that a lot more from them if they were just upfront about it. But no way is that true personalization and they know it. Thanks again Nolan!
Yah I think it's a solid first step and hopefully we get a little more control over it in the future!
"Please do free work for us"? You mean how Midjourney have done free and illegal work for everyone using their platform?
A document leaked with names of over 16,000 artists that they had listed to scrape their work. Artists who have worked hard, created work for both themselves and clients that Midjourney feel it's OK to steal. When will people see the unethical and immoral side of this?
That's really cool. I'm thinking there could be some difference between what I like and appreciate in other people's work versus what I want to be in my own but it's a step in the right direction. And obviously for a different purpose than say /tune or --sref
Hi, thank you for teaching us! I`m IN LOVE with my style.
Yahh? That's awesome. Want to share the code? It'd be cool to check it out
You're in love with YOUR style? So, you're an artist? You developed a style and create artwork yourself? Or are you a "prompter"? Someone who uses programs like midjourney to scrape the work of thousands of artists WITHOUT CONSENT to frankenstein together the images it gives you?
thank you so so much for always teaching us these cool features, you're amazing :)
And thank you for taking the time to comment!
Amazing thank u for tip and how to use and develop personalization 🙏 finally mid journey takes time but they def deliver ❤
Cheers!!
Midjourney is great but lagging behind. We need better coherency, control net features, better text. Use it to generate ideas for work and lately get better results with ideogram. These updates are nice but seem shallow. For a paid service i expect more. Anyone else feel the same?
Creating multiple images with slight alterations would be perfect for selling efts 🤔
Great explanation, can't wait to try it this evening 👍
Cheers!
Top man, great overview
Cheers!
Aesome man! Thanks❤❤❤
Cheers pal!
Great video. I'm not sure what to think about this new update, maybe the best thing to do is to use it and see if it's useful or not. I think it's one of those things that some people come to love while others hate it with all their heart haha.
Thanks for the overview.
Cheers!
Thanks for the video! I was sceptical but when I tried it just with the promt "flowers" and some different values for style and chaos, the results had been impressing. It was interesting that the styles and types of flowers were very different, but always in pastel shades. Although I wouldn't say that this is my preferred color scheme. That surprised me, but the results were quite pretty and definitely usable
Thanks for sharing your experience! I wonder how many other people will be 'pleasantly surprised' with their own results!
My question for the MJ Team is whether it would be possible to reset the personal preference. Or develop more than one. Having said this I'm going to give it a Go and see if I like my preference ;) Thanx for the info/update ;)
I've heard other people ask for similar features - so I think we might see something like that in the future!
As of right now it is stated: "There is no mechanism by which you can reset, clear, or restart your ranking. Just keep ranking and your style will soon converge on your new preferences.". So I would imagine those who have a very high number of images they've already ranked, are gonna have the toughest time depending how many of those images they ranked do not fit their personal preference.
Yep - I have totally different aesthetic preferences for illustration vs photography.
Dont you see what midjourney is doing with this feature? they want as many people to rank the best looking images to get as much new data to train V7. Very smart move
maybe!
I understand the reasoning but I don't see how you conclude this.The images in rank images are often the same, in the same styles. Furthermore there aren't even any images made in Niji.
They Definitely doing that no doubt 🎉
It's better than the alternative if they want to stay out of the courtroom over copyright issues.
Absolute nonsense
tried it yesterday, kinda cool, but you really have to be selective with the image you select in the ranking and explore gallery to really have the personal style you want, sadly the explore and rankings galery are kinda limited
Yah I wonder how much more control we'll have in the future
Thanks once again for an insightful Midjourney exploration. I usually rank images to get free fast hours and now will be a bit more selective (rejecting many) as now it will affect my style code. Wish the rankings would compare like images so you could select the best one of two. I realize this is a work in progress, but if I like my personal code, hope I can save/archive the code for the future-like rerunning a prompt. I will be interested to see if the code changes as I continue to rank images. Great job!👍🏻
Why do you suppose they're setting "Your Style" to the match other people's generation that you like, instead of generations of your own that you like? I often enjoy other people's work, but I don't want my stuff to look like theirs. Thanks for all your great work.
there are a gazillion styles Midjourney uses. with these picks, you teach Midjourney wich of them to generate for you. like a million styles that you prefer instead of the other trillion others. you need to manually prefer one over the other.
Can always count on a great summary here. Thanks, Nolan! Alright, I was never the most interested in this feature, but at least something new has been released so I don't get to whine this week that nothing's come out of MJ :)
😂 I hear you!
Theory: ar 16:9 = Landscape, so midjourney finds a contradictión between "portrait" of a penguin, and "landscape" of the parameter, so it cancel both arguments (in parameter terms, not aspect ratio). Great video, by the way!!!
Well thanks for your insight!
What I don't quite get is why they aren't factoring in the likes from your own generated images. I feel like this would be FAR more accurate in giving you a personalized style. I don't generally ever rate images on the site, and while I have about 2k ranked pairs, it's only for the sake of helping to train the bot, and I generally never really "like" 90% of the stuff I choose.
I used to never use the "like" feature on my own images, but since I've swapped to using the website, and since they no longer give you the standard discord upscale (which isn't an upscale at all) for images you want to keep from a set, I've went to liking all of my preferred images, and that's a ton by this point. I do that to be able to filter out and only see my best stuff, but it would be great for this personalization feature. Complete fail IMO.
Also, this seems kind of a weak feature overall. It's really only of use if you don't want to prompt for style or use a style reference, because those things massively override it to the point it seems quite useless. It's kind of just a "curiosity" feature that you try out and then most likely never really use. I could be wrong about that for some users I guess, but feels like they've sunk a lot of effort and time into this that should have been spent on more important stuff.
I think it's a feature they wanted so that users can drift away from the 'standard midjourney look' - but yah, some more control over the outcome would be pretty cool!
I'm relatively certain it already does. Especially with cref. If you vary subtle on cref images and like them as they get better then next time you run the same cref the results are more like what you liked. In my experience.
That would make sense of they were actually trying to help customers as opposed to just getting people to rank after they ran people off by temporarily taking away the free hour thing. A lot of people never came back to ranking after having to learn how to deal. I'm one of them. And they keep making choices that serve them even if users ask for other stuff. I keep quitting and restarting and I'm sure that loses them money for people who are now on the fence and trying other stuff vs. those of us who used to be 100%.
If you ask for a portrait with a portrait ratio..makes sense. Ask for a portrait with a landscape ratio..less common so maybe it gets confused..you might be better off asking for a painting.
DavidH just gave a big middle finger to users who were spamming image ratings just to get a free hour. Their --p will forever be useless. Serves ‘em right!
The dogs with the snakes 😂 love it ❤
😂 me too
OMG I just tried it and they came out beautiful 😍
That's what I've been trying to tell you!!!
So many good niji images and I have to throw most away... Hands please midjourney... Hands!
haha yeah I heard Niji is a little wonky recently
The painter effect disappears because you are using an AR common for cinematics. So it will have more cinematic source material which will affect the weighting - would be my guess, I actually have no idea
Thank you!
And thank you for taking the time to comment!
My problem with --p is I bounce around all over the artistic landscape, one personalization isn't going to work. I'd like to see Midjourney add the ability to make multiple personalizations that can be fine tuned. For example: starting a --p B&Wtoon by feeding it all of the line drawings I like, then making a --p Colortoon that contains completely different styles. So here's hoping they come out with that.
awesome 💪
Yeeee
If this next update doesn’t fix hands, why are we here?
great update - defintly when I use --p imagies are better:)
I'm glad to hear that!
Impressive work! It’s similar to the AI art customization I’ve seen from CGDream
Cool!
Thank you Nolan
Cheers pal!
The AR thing might be because ai models are trained at a fixed aspect ratio. MJ probably switches to a different model when you change AR.
That's really interesting!
Can we create multiple personalization codes? I can’t see myself using only one personalization code. Maybe I might have a main style that I prefer, but I would probably want to create multiple personalization codes to switch between different styles.
im gonna try it out for a month and see how it compares to sdxl :-) cant wait to try it!
Enjoy!!
Finally something new. I'll wait until v6.5 to jump back in and sub though
Thanks!
Cheers!!
If you carefully select the ranked images that fit your style, --p becomes extremely useful for creating quick and simple prompts with excellent results.
That's a good way of putting it!
MJ releasing model personalization today. Get at least 200 image ratings and then add --p after your prompt. Personalization completely changes the way system interprets your prompts and MJ think you'll really like the new direction it takes. Have fun! ❤🎉
Nice!
💯
Thanks for the update! Is there a way to make the parameter like a vanity url on TH-cam?
at the moment it is pretty random, but maybe i want it to be --p hedgehog
hahah no but I do like that idea!!
pretty cool
💯
Cool feature. While I don't use Midjourney, it's cool seeing lots of the top AI art platforms making new features to set higher standards for both quality of life and desirable image output.
I agree with that sentiment!
Big question: Is our preference profile determined by our ❤’s and top rated generations or exclusively by choices we make on the 1 vs 1 comparisons?
Not our top rated or any of our own images - but yes, images we like AND the 1v1 comparisons
Great work as usual, Nolan- dumb question but what's the difference between SREF and Seed #?
Seeds are a starting blueprint ... same prompt + same seed = same picture (99% of the time) ... sref is a way to point Midjourney at a style you want! You can use '--sref random' and MJ will generate a 'code' for you - it will be a random # but it isn't the same thing as the seed #
hope that helps a bit
@@FutureTechPilot Thanks, Nolan!
I don't really get the point of ranking images. Most of the time, i get 2 completely disparate images and my choice between the 2 is pretty arbitrary
I think that's the point! If you do that enough times, the bot will learn things about you that you may not have realized - and you also get free fast hours for participating in the rankings!
Yup!
They should really work on dynamic poses. It would be so cool to actually create an action comic with midjourney. Because currently, if you ask for an action scene then, it’s just a character posing in front of the camera with no enemy sighs
Good video. I personally think this feature is garbage after using it myself and after seeing your experiments.
My guess is that they decided to work on this feature not because it's actually especially good or useful, but because they want to incentivize people to rate more and this was a good way to do that.
I've rated 2259 images, but I find I really don't like the images it gives me when adding my personalisation code :( But, I'll keep rating images, hopefully it improves. Very informative videos, really enjoying them, thanks.
I think it would be cool if people could 'reset' their ratings, now that they know what it will be used for
I thought a personalize feature I would have more control over what I’m making but this is ok
Love the images that come out of mid-journey but it's the most unintuitive experience out there.
😂 I think they realize that and they're trying to work on it
A question: Ranking has been with us for a long, long time. Even prior to V5.0 if I recall. Has MJ remembered all our preferred images from then or does it only apply since this feature has been released?
That's a good question! I don't have the answer for you, but I *think* it's all the images we've ever ranked (maybe the weights are skewed more to recent images or something like that)
The aspect ratio is very connected to style. Portraits are typically "painted" in historical times in a portrait orientation. Wider portraits are a fairly new invention for photography. Just like panoramic images would look more cinematic for a movie for instance. Seems like common sense to me. If you pick a square portrait you might get portraits that look like they were taken with a Polaroid.
Nope... portrait photography was horizontal BEFORE It was vertical.
@@_gabrielln.dop_painted portraits were WAY before photography
Nice overview. I wish you could train on your own images.
Maybe this type of feature could lead to that one day!
@@FutureTechPilot I guess sref is sorta like that already.
feels like a very specific --sref basically. Wish you could just make your own models already
Yah definitely feels like an --sref!
Is it just a replacement of style reference? I wonder what will happen if you put a different style reference with your personal style.
That sounds pretty interesting!
i feel like we should be able to do folders like pinterest based on styles etc, then refer to these instead of a generalisation
That sounds like a brilliant idea! Maybe consider leaving that message in one of the feedback channels on Discord!
I don’t understand why this takes reference from rating images instead of images that you generate yourself.
Are the images you like have a better vote weight than the ranking task?
I don't *think* so - but I guess I don't really know
my usual {slightly} negative take on this - great video and explanation btw... is...
what's the point, really?
wouldn't it make more sense to have 'determined' what a person's preference is based on the images they make rather than those they rank?
Using MJ to make tens of thousands of images (because nothing comes quickly) for different projects with different aesthetics... being given one specific 'style' seems at best redundant at worst, pointless.
I get it - it's a fun game to play with for a couple of minutes - but it just feels like it's part of the dumbing down of a potentially useful tool... couldn't someone suggest they actually put their time into things that would either benefit everyone on MJ or make it a more useful tool... such as things people have actually asked for on the ideas ranking list... like v7, text to video, 3D creation, better moderation, stuff like that.
They take all the rankings or just recent ones? maybe last 200 ? How it's possible to get something I like from 50k images. Would love to have option where code is created only from photos I liked on Expole page.
Yah I hope there is a little more nuance to it in the future - because like you said, some people ranked images without knowing this would be a thing!
@@FutureTechPilot Think it's just a trick to make people use ranking tab more . Hard to make something of your taste when you chose from 2 random photos, it just feels like some random style. Better just make code from photo you liked(hearted).
Uh oh, I didn’t always pay attention to style, sometimes I just liked for tech accuracy 😮
haha now you know!
That´s some weird and amazing new feature! I just realized I have rated 17777 images until now. omg when did i do this =)) - whoaaaaah and the resulsts when comparing without and with --p... holy sh... :D
Excellent demonstration. Thank you for sharing this new trick. But please avoid showing snakes in your video. 😱😫😟
haha okay okay
@@FutureTechPilot Thank you handsome. 😘
They should've used the Left Arrow and Right Arrow keyboard keys to rank images. Duh.
haha yah I guess that makes sense!
So cool, can you create more than 1 code? 🇦🇺❤️
cool i guess, i mean its not that different from the style codes. It's also obvious that it's just a way to get people to train future models 😆. People really just want to train on their own images
Can different codes be developed for different preferred styles?
That's definitely something that has been suggested - we'll have to see what they are able to come up with!
To me it is more of a photoshop filter or preset. For me I just got a TEAL cool filter on my images.
I'm not interested in having fun. I'm interested in evaluating the commercial value of it as a tool for graphic design work. So far I'm not very impressed. VARA compliance needs to be coded into the system. Prompt text that mentions an artist by name can be challenged by artists covered under the act as a violation of their moral rights if their name is used on someone else's work. What happens if a Midjourney prompt with an artist's name in it manages to get a higher search engine ranking than the artist's website?
I dont get the point in using the seed number. can anyone please help me understand?
It will just help you recreate the same grid of pictures if you use the same prompt and same seed # - th-cam.com/video/rpoSYuuYrs4/w-d-xo.html - check out that video if you want to learn what else you can do with them
Is it like training a personalized lora or checkpoint?
No unfortunately. But maybe we'll get something like that in the future
11:53 I wonder how Midjourney will apply the personal codes to Niji Journey.
You know, I didn't see the real advantage over for example --ref. My opinion at least.
You're not wrong! I think this way might just feel a little more 'personal'
I don't see how it's useful, given I switch styles & themes constantly. Perhaps it would be nice to setup some standard sets.
Yah I agree. My code works really well for people but I don't like the results it generates for anime/cartoon
I’m very interested in this stuff, but for some reason the more I watched the more bored I got
😂 I'll try better next time
How is this better than just using the style parameter?
Not necessarily better, just different! More personal to your tastes
My problem with midjourney is it has no clue how to do spaceships, not a single one is usable, I think there's more weight to the complaint that the Midjourney AI is just mashing things together than making anything new. I tried to give it super detailed and accurate descriptions for space ships n it never worked, even got ChatGPT involved n it never worked
Does this personalization feature brings you talent? Or does it make you feel like a real artist a bit more?
😂 it's just your personal preference, it's not that deep
You like dogd with glasses. XD
😂 I'm a simple man
👋
🤝
Wide aspect is more cinematic i guess, built in bias
That's a good point!
Trust the "system" ... oh sure, sure... that's exactly what they wantvyou to do!! 🤣🤣🤣
😂 Yah maybe don't listen to me on that one
Why --seed though?
to lock in the most random variable so we can test the effect of other changes
What is--seed ??
Just a way to reduced the amount of random variables. If you run the same prompt twice, you'll get two different seeds = different pictures. If you run the same prompt twice on --seed ### = the same pictures
@@FutureTechPilot thank you lot your video is really healpful
Personally I think this update is quite useless. basically this is another style reference comand but poorly implemented. I search a for a style only 20% of the time i use midjourney and only when i'm chilling or experimenting But for 80% of the time, i try to generate WHAT I NEED, and that si still something difficult to get right. the guys at midjourney should focus on other problems that midjourney has because as much as progress has been made, we are still far behind in terms of the usefulness and practicality of midjourney.
I like exploring the default midjourney too!
Will it ever be free? 😅. Like Leonardo😰
impossible bro, they are a business. no businesses go from paid to free.
There might be a free tier one day! (they used to have one, but it was abused) - maybe they'll reintroduce it in the future
Leonardo will also be for-pay eventually or disappear. Generative AI is very very expensive
I just discovered that Midjourney has gone way downhill in reproducing STYLE (Wes Anderson style, for example). It was brilliant in 5.1 and 5.2. Now it is terrible. I canceled my subscription. Let me know when they fix this.
haha I noticed that too! but I bet there are ways to prompt for it
pointless feature. they were talking about making realtime graphics and game engine type stuff at the beginning of the year...
yah that's where the tech is going, not where it is now
Yawn
wow MJ is not only too expensive but too greedy. Having us pay to train their models is just evil
the guy promoting shamelessly plagiarism ..
I hope you have a good day pal
Another useless feature, with no control whatsoever. They better make video soon or.im.out