Mastering Midjourney v5.2 : A Deep Dive into the New Version! (Zoom, Vary, Shorten, Tips & Tricks!)
ฝัง
- เผยแพร่เมื่อ 31 พ.ค. 2024
- In this video, we explore the latest update to Midjourney, V5.2, and its exciting new features. We compare V5.2's more stylistic approach to image generations and the most exciting feature of all: Zoom! Plus, we’re going to dive into Midjourney’s most important (in my opinion) new feature: Shorten, which gives us a REAL look into Midjourney’s Language Model. I’ll also be covering a few really interesting tricks and techniques that you can use with 5.2 !
Links
Midjourney Prompting Video: • A Perfect Midjourney P...
Join the Discord: / discord
AFFILATE LINKS:
Camera: amzn.to/3yXMDY2
Microphone: amzn.to/3K1jIZm
Audio Interface: amzn.to/3lDX9kf
Coffee: amzn.to/3JZuBeq
Follow Me on Twitter: / theomediaai
-------------------------------------------------
Thanks for watching Theoretically Media! I cover a large range of topics here on the Creative AI space, Technology, Tutorials, and Reviews! Please enjoy your time here, and subscribe!
Your comments mean a LOT to me, and I read and try to respond to every one of them, so please do drop any thoughts, suggestions, questions, or topic requests! - วิทยาศาสตร์และเทคโนโลยี
I have to tell you I took studio art in High School and had a passion for art. Years went by and that part of me faded away. With all this new technolgy it awoken my passion again and I'm beyond addicted!! The fact that you can combine your skills with this "Magic" is mind blowing!
I love hearing that! It’s funny, one of my Midjourney workshop ideas is to take a memory from your childhood, or an old photograph, and use that to inspire a piece in MJ, I think it would be such an interesting idea to reconnect with your childhood. It might even have some art therapy applications as well- although, I’m not trained enough in that field to make that call!
There are a lot of really interesting ways to use /shorten
Something I've been doing is to run images through describe and then feeding the four results into /shorten. Create an intelligent prompt from the words that are not crossed out. Run THAT through shorten, then use not crossed out words again to build something intelligent. The results are... pretty outstanding.
Thanks for the breakdown! Concise and comprehensive!
I had actually meant to do that describe/shorten experiment, but the video was running long already. I’ll save that for next week! It’s awesome: I’m excited about Midjourney again! (When I try to leave, they pull me back in!!)
Thats Ace. Thx for sharing!
@@TheoreticallyMediaexactly what I am thinking 😅 They keep us addicted
I’m so interested to try this now.
@@airdrummer55 Give it a shot, it is a LOT of fun! Plenty of stuff on the channel to get you started! This is a pretty good starter: th-cam.com/video/j-GZvXgfgaY/w-d-xo.html And feel free to swing by the discord if you have any questions! (or ask here!)
Man I went from zero knowledge to somewhat decent only after 3 videos. You have great skills in teaching!
Oh man, thank you SO much! Seriously means a lot!
Super impressive, I've had a lot of fun using the zoom out feature although it gets pretty funny after about 5 zooms! The shorten command got me excited as well while trying it but the results have varied - some images are very similar and sometimes not. Awesome update from the team, and thanks for the video too!
Thank you so much for taking the time to break down these things and help us newbies to the wonderful world of Midjourney. Much appreciated, keep up the good work! Saludos from Cambodia!
would never have tried Shorten, didn't know about the weighted token summary! this is gold, I'll never write 4K again!
Ha! Not gonna lie, I kinda forgot about it too! Actually, from your comment, I’ve now got a video on the “to do list” that is basically: The Forgotten Midjourney Commands!
Thanks for featuring my zoom out video ❤
It was soooooo perfectly demented! I LOVED It! It was actually one of the first tweets of the day that I saw, and not only was it the best of the day, it also made my week!
Great breakdowns. Love your channel!
Thank you so much!!
I really appreciate that you went for a Danish Instagram model 👌🏻Thanks 🇩🇰
Thank you for taking your time to do this video.
Keep on with the good work :)
Thank you for watching! It was a bit of hustle to get it done, but I'm glad I got it out into the world today!
Subscribed. Great flow. Easier to follow along, pick up on new concepts, and fun. Thanks man!
That you so much!!
Just watched two of your videos after finding this from browsing. So well presented, easy to understand and inspiring, actually. Cool.
Oh, thank you so much! Always happy to hear the I-word!!
Delightful informative as always
Enjoyed the video, this helped explain a lot about the new stuff. Enjoyed the background music too, compliments the voice over
Thank you very much! I actually write and compose all that stuff myself! It's totally a highlight of doing videos and is very much my "zen" time of the week! Much appreciated!
Nice overview man! I feel like your channel will pop off soon with the rising popularity of midjourney
Thanks Tim, excellent info !
Thank you so much!! Hopefully I'll be able to get to the dash dash command list video next week!
Played w 5.2 today for a few hours and LOVIN 💯🎉
Great content!
Thank you!!
Thanks Tim, a very useful review. I agree, zoom will help with character consistency. From a few rolls it does seem more realistic, even when going for watercolor images. Anyway, thanks again!
Fantastic thanks Tim
Thank you!!
Fantastic video, thank you! 👍
Yes, every one of the new features is pretty darned good, and the 'shorten details' is our Diploma for better prompt engineering!
I have noticed that everything leans towards being pretty photorealistic. That's great, _if_ you're looking for photorealism. The market defines the niche I suppose.
amazing video, thanks!!
Thank you thank you!!
Leonardo uses about 12 different fine tuning models to generate the prompts you give it. You can literally take one prompt and get 12 different interpretations. I’m not subscribed to Leonardo like I am with MidJourney but what I have done using the free subscription of Leonardo is if I come across a generation that I feel is ideal for a project, I take that generated illustration over to MidJourney and use /describe first. Once I read what MidJourney has interpreted, I may select one and change little things here and there ex: aspect ratio , camera shot etc. Hope this helps anyone . Have a good day!
Good tip! I'm going to be spending some time in Leo next week. It's been a bit since I've really dug in, and they've done some pretty great stuff in the meantime. Looking forward to it!
TQ Tim for this wonderful vid. Can u consider to do a video about consistent character prompting tips with V5.2? Thank you
ok im sold with the outpainting. thats been one of my biggest gripes with ai images.
It's a MASSIVE gamechanger for MJ, in my opinion. And apparently they ARE working on Inpainting...v6, perhaps?
9:00 It always amuses me when Americans of Brits try to pernonce Dutch names...
Ohw well you had a good crack at it 🤣🤣
Great video as always!
Ha! I’ve had a number of Dane pals over the years, I love you guys, although Stimorol Liquorice is terrible!! Haha
That zoom feature is a cheat code. I also read you could zoom in on the zoomed output. Awesome.
I’ll have to try that! I did try to add a negative weight to zoom, but no go there. Someone on twitter had a good tip, to start with a Macro shot and zoom out from there, so basically you always have an extreme close up!
Thanks!
Oh my goodness! Thank you so much!!
Another good review. I'm disappointed that the zoom feature isn't effectively a resolution boost, but the overall image sharpness is much-needed, and /shorten may initially be seen as ho-hum, but I think it will be crucial to perfecting prompts. I look forward to using the insights from /shorten to improve AI-assisted prompt generation.
Oh, I’m going to be playing with shorten all weekend! It’s funny that I just did a prompting video, because now I’ll probably have to basically revise the whole thing!
Can’t wait to see you dive into Leonardo AI! Would also be interested in seeing you play around with Playground AI as well.
Thx for the great info, much appreciated.
Is there a relationship between 'high/low variation mode' & 'chaos' parameter?
Looking forward to the Leonardo deep dive!
Hi, thank you for your video, fantastic as always, did you already saw the Midjourney animation option?
Really good vid...thanks! I especially like the part about MJ providing stats regarding how it recognizes words and provides weights for individual words. Just out of curiosity, do you know of a "list" of words (I know, it could be thousands of them) where we can identify those that MJ recognizes the most and assigns the highest weights? For instance, I like playing the fashion photography genre. Are there certain words, with weights that MJ recognizes more than others?
So, sort of? I go over the shorten command in this video, which, I think if you pair with Describe gets you pretty in the zone! Midjourney's Newest Weird Feature (Deep Dive & Tips/Tricks!)
th-cam.com/video/R6lJJAZ2GQc/w-d-xo.html
Great content, I really like the new features. But your flag emoji tip blew my mind. I had no idea 😅
So crazy, right?!
why adding --q2 if it's 1 max atm? Or is it the older ver?
Great info! When I hit the 'vary' it automatically changes the image i.e. I don't get the box to change anything (as you changed to 'forest atmosphere'). What am I doing wrong?
Did you upscale one of the initial 4 grid generations? Vary only works on upscaled images.
@@TheoreticallyMedia Yes I did.
Not understanding how you put in the city image behind the bike with very(Strong) Can you elaborate how to do that and also how to put in forest atmosphere behind the bike? Thanks and really love your videos.
100% So when you run a Vary, you have a chance to change the prompt. In that case, I just added "In a City Atmosphere" to the end of the prompt, and then ran it. Sorry, that might have gone by too quick in the video!
Very cool! How did you get your face on Superman body? I didn’t know we could upload. Thanks!
that's a really interesting Midjourney plugin called InsightFaceSwapper. You can check it out here: th-cam.com/video/PPQl6qEr5Es/w-d-xo.html
The shorten is very interesting. It seems to ignore much from my long ChatGPT prompts
Cool man. Cool.
So chill…
I find it pretty funny to run describe, then shorten on the prompts. I’m doing some experiments with it, chopping out any of the strikeout words and using the remaining words for a prompt and comparing them to describe prompt results.
So, I don't know if you saw it, but I found a pretty crazy/fun method utilizing /describe and ChatGPT-- I've never been a big fan of using ChatGPT to write prompts, since...well, it isn't as fun? But this ended up being kinda cool: th-cam.com/video/7ttBvM8i9_o/w-d-xo.html
@@TheoreticallyMedia Ooh, yes, I did see this video! It was great!
what is the prompt formula for the cover of this video? thanks
I actually don't remember 100%, but for the MJ stuff I usually use something along the lines of "Stunning" followed by a keyword from the video. It's really simple! So, for something like this one, it might have been "Stunning, Dive" -- I mean, honestly: It is that simple! ha! Hope it helps!
Best new features - Zoom, Strong/mild variation, and **shorten** 🤯
Shorten singlehandedly is making all self-described 'prompt engineers' have to rethink all their prompts that weren't actually affecting anything.
Only thing I disagree with on shorten is if you have 5 pictures they are definitely blending. It claims that it had 0% affect on like 4 pictures after the first one, but for sure in 5.1 I included both fire and the irish flag in my pic and the characters were coming out with flames and irish color scheme, so I don't know.
The zoom out is so wild though. It's like if it's magic. I also have a subscription to Gigapixel AI (for upscaling further) so I've basically been creating wallpapers all day.
I am addicted.
Oh, I'm gonna be playing with it ALL weekend! I don't consider myself a prompt guru (ugh, they drive me crazy)-- but I love the fact that you can run one of their bewildering prompts and call them on it now!
I'll look into that blending thing you mentioned-- I haven't gotten enough time to play with it, so that'll be part of my experiments over the weekend!
LOVE THE ZOOM!!!!
Is it possible to make them in a 360 or 180 degree format to view in VR?
not natively in Midjourney, but there are some Skybox tools that I'm looking to cover soon that will do just that.
hello Sir, can you talk more about the available face swapping method to combine context changes with the Midjourney application? I see you can also combine other photo creation applications. he is very good. hope a response. I follow you from the other side of the world. ^^ . Thank you
Think of the improvements from each previous version of MidJourney to the next, and what we've got even with just from 5.0 to 5.1 to 5.2 ... Can you imagine what version 6 will be like? They might've even figured out repeat characters by then!
Could you pretty please do a video on how to blend the zoom outs like @lemoonsynth did? The feature is blowing my mind, but I can't figure out have to transition between the images like he did.
It almost feels like video generating the zoom outs. The way one sees consecutive zoom outs is really suggestive of a narrative taking place.
Watching LeMoonSynth's video a bunch of times, I think there was some tricky stuff going on-- If you watch the full version on twitter, it zooms out, then back in-- but there are changes on the in. Perhaps we can use the power of the AT to summon LeMoon and they can shed some light? Would you be willing to share some wisdom @lemoonsynthography4698 ?
we need a video about Leonardo's new update: Alchemy
I agree!!
do you know how to make that Zoom animation effct?
There are a few different techniques. I’ll try to go over them in a video soon!
i'm confused on how you change the prompt on a Vary Command, I can't seem to do that
Did you upscale the image from the 4 grids initial generation? Vary only works on full upscales.
@@TheoreticallyMedia yes, i have the Vary options, both strong and subtle, but i don't know how to change the prompt. When I'm presented with the options i can only press the buttons, it doesn't give me an opportunity to change the prompt like it does if i hit custom zoom
@@DisentDesign Oooh, I know why: Go into /settings and make sure the remix option is turned on. That should solve it!
Nice Superman photo 😂
Ha! I don't know how that started up! I think it was in the initial InsightFace video-- I'm much more a Batman guy, but that doesn't really show off Inswapper, as half the face is covered!
The /shorten command lies by omission. Generate an image with 'by ' with an artist with a distinct style; say Jackson Pollock. There should be some evidence of Pollock's style. The same prompt with the /shorten command will provide 4 options whit no Artist Name and the artist name will not be in the tokens. In your example 6 artist are named in Leonard AI prompt you use they are ignored.
A quick test with specific styles shows that they are not being picked up by /shorten . A prompt which includes water color sketch will produce a water color image however the shorten command will drop the 'color sketch' words in its response and the options it provides generate standard midjourney images. So style parameters needed to be added back along with the shortened subject prompt words.
😮 we're already at 5.2?!???! We'll be at level 12 by the end of the year at this rate lol. But honestly I've been sticking with v4 lately cuz the 5 and 5.1 results have just not been reliable. Before 5.1 rolled out, v5 was spectacular. But now I can't bother with how often the 5 series has been getting misaligned results, watermarks, cartoon results wasting my credits, and several with too many artifacts and physical "deformities" lol. It's like when Windows had XP which no one wanted to move on from for Vista or beyond.. I'll hold out in v4 for some time til the updates are truly worth it
I mentioned that in a video awhile back, where I was trying to do photos from the early 90s grunge era-- like, when Nirvana was playing clubs. The photographer, Charles Peterson, had such a great energetic and...well, grungy style-- v.4, while not trained on him, does such a good job of capturing that energy. 5 and onward are just a little too clean and sharp.
I mean, thankfully, MJ lets us use all the previous models, so I can't complain!
@@TheoreticallyMedia yes friend, I saw your video on that and was so nostalgic!!! '92 was the year 👏🏽👏🏽👏🏽 yes so thankful we can choose and switch between the versions! Extremely interested in the 5.2 zoom function really perfecting tho! Been having some awesome success with music Gen also and might pay to double my space, I'm that impressed. Going to play with Kaiber this weekend and enjoy exploring it. If AI can streamline and advance music production in the sense of composition, music theory/layering voice/lyrics/melodies/harmonies etc... 👀👀👀👀👀👀!!!! I'm excited and appreciate your work helping us dive deeper. Now... I wonder if MJ can capture various music genres in image, including world music, time periods like 60s Motown or 70s disco, hard rock, etc 🤔🤔🤔🤔 🎵🎶???
@@purplepink5630 That is SUCH a cool idea! I might have to spend some time digging into that idea as well! I did some stuff awhile back on v4 where I took an AI voice generator and had Morgan Freeman do a Punk Rock song from 1977-- you might be (one of the few people) interested in checking out the production video on it: th-cam.com/video/FQg6VY6NsEE/w-d-xo.html
Let me know what you think!
@@TheoreticallyMedia ❤️❤️👏🏽👏🏽👏🏽 will def enjoy checking it out!!! Love this! Yes, see what you come up with and I'll try this weekend also and check in on what came up... I'm thinking say if there was a completely unrelated scene (I e. A pool party) but set to 60s Motown in MJ lol ... Or even a base or story line for Kaiber to generate an evolving video in the style of those genres🤯 lol off to the AI playground we go!!!
But still cant fix the issue with limbs and fingers (is 5.2 possibly worse in this regard?). Midjourney really can't count but, hey, it still is the best out there and getting even better
Yeah, that’s an overall AI problem: I think Vox had a pretty good video on why AI struggles with hands so much. I’ll try to dig it up.
Limbs I feel…yeah, you know what? I have noticed more wonky limbs lately! I kind of just zone them out, but now that you mention it, I have seen that more often than usual…
I have paid accounts in both Leonardo and Midjourney and I actually prefer Midjourney. It’s kind of a pain inside discord but I find I get better results with Midjourney. Maybe personal preference IDK. One thing I don’t love about Leonardo is that it almost always crops initial images too tight, forcing you to burn credits outpainting or export it to PS. Leonardo is obviously better at inpainting in canvas mode but I still prefer Midjourney.
The inpainting and outpainting are two things I utilize Leo for-- but I also really like the Pose-To-Image feature when I'm really aiming for something specific. It's interesting to hybrid both MJ and Leo-- do stuff like posed characters in Leo, and then backgrounds in MJ. I think it's one of those things where its best to look at the strengths of both and try to use them together.
Oh yeah, why do you ask for a blue color palette? I just figured the blues and oranges are default. ?
Oh, y;know, I don't know-- but you're right, that whole Blue/Orange (cold/warm) look is pretty standard. Next time I'll go with something weirder-- like Purple!
I don't think InSite works with discord on the phone
what does --q2 mean?
Quality. So that means how long MJ spends “drawing” an image. 2 is default, but you can crank it up to 4. Just be aware that it burns your hours faster though. Gotta be careful with it. To be honest, I usually just leave it at 2, unless I’m really confident about what I’m aiming for.
@@TheoreticallyMedia thank you for your clarification. 😊
Don't faces seem a little stylized compared to an actual photo now? It's something about the skin texture. It may be smoother, different normals, or a lack of subsurface scattering, idk. To me, the skin texture definitely looks more like what you would see in a 3d model render that isn't detailed at an ultra-realistic pro level.
When you cranked --s 1000, I think some of the images had more realistic skin, but some still had the issues with the texture. idk. maybe it's just me.
Yeah, I hear you. I think we're in a weird interim between v4's "wax figures" and ultra realistic skin. It isn't QUITE there yet, and in v.5x almost seems to have gone into Unreal Engine territory. Don't get me wrong, it looks great, but almost too perfect.
Hi! Your info is great. Much appreciated. New subscribers here. One comment tho. Many of the generations you use as examples look like digital paintings rather than photographs. Digital painting is a wonderful look but if the preference is an actual photo or a movie still rather than digital painting, a human generated prompt detailing the camera angle, camera and lens type, f stop, aperture setting and yes even the film type will produce stunning photograph type generations. Also using a photographers name or better yet a combination of photographers names will evoke an actual photograph. Using the term “photorealistic”WILL OFTEN YEILD POOR RESULTS. So avoid using that term in prompts if a photograph is the desired style. Just a photographers 2 cents with many thousands of generations as proof.
Thanks for the tip!
You should try putting one of your super detailed camera info prompts into the /shorten command and see what gets crossed out
Resolution remains an emergency.
I generally just take my images a use an upscaler. That said, it'd be nice if either that option was built into MJ, or it produced higher res images. I'm sure it is capable of doing so, but I'd imagine the GPU resources for all the users would slag the building!
Is that a Schecter I see on the wall?
Good eye! It is! It’s the C1 E/A with the F holes, I love that guitar. Tend to go with “oddballs” a lot, and not the standard Strat/Les Paul/PRS ones. Those are great too, I just love stuff with weird character.
The one next to it is another oddball: It’s an 80s Peavey, swamp ash (heavy), with this strange system of splitting coils. Turn the tone to 10 and it is a single coil. Roll it back to 7 and it is a humbucker. Guy I bought it off of said it was owned by The Nitty Gritty Dirt Band. It’s such a weird thing to say, so I have to take it as true. I mean, of all the bands?
Anyhow, how about yourself? Are you a Schecter Slinger?
midjourney does still have serious problems when u ask about realistic anime style,if u want a full body shot realistic style it comes with the same in all past versions,bad quality and resolution and also bad face features,in th eway i use it,i dont see any improvement so far :(
Have you tried Niji? I still have to dive into that mode a bit...Let me know if you've tried it out and what kind of results to expect!
👋
Thank you Louis! Good to see you!!
wooow, outpainting! finally.
I am SO happy! Rumor is that in-painting will be coming soon as well! As soon as we get consistent characters (also promised) and something like pose-to-image, MJ will be unstoppable.
I hope they reduce the subscription price, there are competitors
Sadly, I don't see that ever happening. If anything, prices will just go up. Although, as they add new (higher res) outputs, I do see a two tier'd platform model. Something like "Base" and "Pro"--
One thing I'm curious to see when they release the web UI is if they'll do the $10/mo plan with "relax" mode-- Relax doesn't burn hours, so technically, you could have unlimited generations, they'd just be slow. That said, keep an eye out, because when they do launch that, I'm sure they will have some kind of deal going.
It’s 8 dollars
All these features and no type of controlnet or training. Only so pretty the images can get. We need controlnet style consistency
Agreed. It does sound like it’s coming (at least they report so). I’m wondering if we get in painting and some sort of Control Net when they launch the web version.
Admittedly, control net features would be pretty annoying in a Discord interface,
Sorry, but the music in the background is cruel.
Haha, loud? I set it to -26db. Always trying to find the right balance in the mix.
@@TheoreticallyMedia No, not out loud. Just annoying this tootling. Thank you. But there will be reasons that are not clear to me to do that.
LOL.
@@TheoreticallyMedialove the music. Great content. Can’t please everyone 🤷🏽♂️please don’t change anything ✌🏽
@@iwillflyfilms Haha, thank you very much! I wouldn't actually-- The music is something I do myself on the weekends. Making beats and playing mellow guitar/bass over them is kind of my relaxing-mode. Don't get me wrong, I LOVE making the videos, but writing music is kind of my meditative place.
Appreciate it!!
Midjourney is an 8 year old who's discovered it has a talent for drawing breasts, it is all it can do: Midjourney/ imagine a dog with a ball: skinny generic dog with breasts. Midjourney/ imagine a sports car: skinny generic sports car with breasts. Midjourney /imagine a handsome surfer under the ocean coming up for air with longboard: fat guy drowning wearing a business suit and skinny generic dolphin with breasts.
Stylize actually goes from 1 to 1,000. Try it, you'll see.
Oh, that interesting! Documentation states that 100 is the default value, so I guess going lower is basically negative prompting stylize. That might be very handy!!