thank you so much! Had some good laughs with this one, but overall-- really impressed with --cref! There's a TON more to uncover here as well! I'm really looking forward to seeing what everyone discovers with it!
Great video. Can you detail your prompts for a character reference sheet? I always have problems generating them. I always need poses front view, 3/4 view, profile view and rear view. I take these views and construct "puppets" in Adobe Character animator. Thanks !
That’s a great usage! To be honest, I usually do “character turnaround sheet, details of character, character model sheet” I think that repetition helps. Although, I’ll say that Jenna Ortega model was cherry picked. I had a lot of bunk characters and incorrect model sheets before landing on her!
lol 🙂 but only *after* i took the fall for that busted casino heist. the blue business suit gang just isn't the same any more since we let open shirt guy go...
On my Midjourney web interface I don’t have a person or paper clip, just the right corner image picon for reference clip. I’m on an iPad if there is some kind of difference? Bye the way great video tutorials.
Hmmm, try refreshing? The iPad part shouldn’t matter? Or, possibly try downloading chrome for the iPad. It did just roll out, so my guess is that you need to refresh the browser.
And it doesnt work. My prompt was : looking to his left, --cref "URL" --cw 100 --style raw --s 250 The result was nothing like the image in the url!! I just wanted the same character but looking left couldnt even do that!
Some ideas: Try using your initial image as an image reference, and then with the prompt. OR, try inpainting your character's face. The inpainting idea is probably the way to go, as I think about it.
@@TheoreticallyMedia Thanks very much. For the inpainting are you referring to using impainting on the original or on the result? I'm also not sure on the prompt for the inpainting would I need to use both and image reference and --cref?
awesome! I appreciate you actually looking at the midjourney discord details about the new features so you can explain them more in depth - Ive watched some videos where they don't really understand how to use a new feature because they haven't looked at that. 👌🏻🔥
oh, that's ALWAYS something I study when making a video. I have a lot of respect for the MJ team, and I try (as much as I can) to get everything right. I know there are some other channels that will rush to get a video out so they can be the "first"-- but something I've always prided myself on is taking the time to dig in and really figure out how a new feature works. haha, that's probably why the Permutations video took me so long! Stupid curly brackets!
Haha. We were probably working at the same time! I’ll admit, my brain was fried out pretty hard yesterday, so I ended up ditching out and doing all my research this morning! Good night of sleep is a great reset!
Can't wait to pair this with Sora's ability to turn stills into (prompted!) videos. That is getting damned close to actual cinema. Gotta see how much control Sora actually gives us in terms of "directing" a character from a still image. I mean, if you gave that warrior gal a gun, could Sora have her actually fire it?
Yeah, I wonder that as well-- I'm still not 100% sold on Sora. The videos we've seen are impressive, but how many HAVEN'T we seen? Also-- did you hear that it takes an hour to generate one Sora video? And that's just the limited amount of people who have access to it. I still say, we aren't seeing Sora anytime soon. We'll see Sora level quality from somewhere soon, but not from Open AI.
@@TheoreticallyMedia Sora is one OpenAI technology I really wish *was* open source. Even if a minute of video took three hours to generate on an RTX 4090, all I'd say is, "do you have any idea how expensive and time-consuming it is to get a minute of footage for a feature film?" I noted another video showing how Midjourney style references can be combined with character references to produce images with the same style and color grading of something like Fury Road (or whatever film look you are going for). I mean, that is tens or even hundreds of thousands of dollars saved. I intend to build my own roster of named vactors/vactresses and use them in one project after another. In a few years they will be able to give their own Zoom interviews to Vanity Fair.
grain of salt as alot is only rumors but ive heard 1/100 cherry picked and up to 60 minutes wait time for 1 minute of generation...on their GPUs. Im also not sold on it being anything new or novel. Really just seems like they released it to keep the hype going when other companies released their AI @@TheoreticallyMedia
ai SFC gonna be huge, id pay $100 per min.6k for a customized movie is insane and going to be an option someday, then ai video games hopefully in my lifetime! @@JohnSmith762A11B
hello bruh currently im working on a movie , script is done by me. i want to create chrachters for my movie how i done thhis ? plsss bruh hlp plssss.. remember bruh i havnt money to spend midjrny
Great alternative is Leonardo.ai. While not free, they do offer free daily credits to get the job done. So, you’ll be able to do it, it’ll just be a bit slower to get there- but, free!
Haha, you should see me trying to get through one of these videos! I'm cracking myself up here all the time! Looking forward to hearing your thoughts on Kaiber! It's such a sleeper in video gen! Love it!
Fantastic, excellent low down Tim! Can't wait to see how this improves and when we can use actual pics not generated in MJ. Also wha happen to Dark Mode, you've gone psycho? jk! ;)
Excellent to hear! There's one thing I didn't hit in this video, and that's to play a LOT with the various Aspect Ratios-- now that we can pan/zoom, a good way of getting a full body shot is to start in 9:16 and expand out from there. There's so much potential in MJ now!
One of the devs (sorry, blanking on his name) will occasionally post them on Twitter. That’s the only current source of Sora footage. He does appear on the MKBHD podcast, where he was asked point blank: “where do we get Sora?” And his answer was: “Not anytime soon.” Also, I didn’t mention it in the video, but one Sora video takes an hour to render! So, yeah: it’s not happening on a public scale for quite some time!’
It can!! You might want to try Niji as well- I don’t cover that much here, but you can get some cool results from it, and even images that aren’t totally anime.
@@TheoreticallyMediaHere's a really cool video, Tim, from a guy who is using the new consistent character feature, image references, and Photoshop to make really good children's books! ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-TxzOiCjfq9M.html
Documentation states no-- BUT, a few people have tested objects and it seems to be doing it. So, I'd say...maybe? Likely depending on exactly what you're trying to cref.
Come to think of it, are WE even real? Or also some AI generated data from a prompt by some guy in another universe? Just wanted to throw that out there following your similar deep thoughts about the man in the cafe thinking the same thing.
oh, I've gone down that rabbit hole MANY times! Turns out at the end of the tunnel, there's Ice Cream. So I eat the ice cream and I totally forget about all my existential dread! Ice Cream solves EVERYTHING!
haha, that's not my department! Sorry that has been an issue for you though! I know I've seen people complaining about that in the Discord, so it does seem to be an ongoing issue.
oh, seriously. Between this, and HeyGen/ElevenLabs/EmoTalker, etc? Yeah...its not going to go well for style based influencers. That said, people sometimes ask if I'm worried about, and I'm really not-- since videos here are more knowledge based-- and let's face it, no one is tuning in for my looks (ha!). Plus, you'll know when I'm AI generated when I stop saying "ummm"...
One of my first thoughts was also about using real people's images to put them in whatever situation. Did you try it with just the one cref image of yourself? Maybe if you had 3 or more images of the same person, but a bit different angles, that might work much better. And as I'm typing this, I'm wondering why I'm not just trying it myself. ;)
Haha. Give it a go and see! But, I do think there is some sorcery in the code that detects real photos over synthetic ones. If you look closely as an image is generated, you’ll see the “real” face appear at some point, but a step or two after, MJ will scramble it.
I hope Midjourney also adds: *1.* The ability to copy a pose from an image, and put our consistent character in that pose *2.* The ability to pose multiple consistent characters together in an image *3.* Consistent backgrounds *4.* Consistent objects (e.g., Sometimes we need our character interacting with a specific object that has a unique look to it. For example, it could be a unique-looking music box, or a sword with a unique design, or a magical object) These things would really complement the character consistency feature! Together they would make storytelling a lot easier. *EDIT:* I added number 4 to the list, since I think that’s also an essential feature for storytelling.
Haha. I totally agree but also always have to laugh, it’s like everytime MJ releases something we’re all like: “awesome! Now all we need is this, and that, and this. And one more thing!” I’m just as guilty! We’ll never be happy, and that’s ok!
@@TheoreticallyMedia Also, I would add “consistent objects” to the list of essential features. Sometimes we need our character interacting with a specific object that has a unique look to it. For example, it could be a unique-looking music box, or a sword with a unique design, or a magical object.
@@TheoreticallyMedia If I were to use Midjourney for long storytelling projects, such as to create movies or graphic novels, I would need to use all of the listed features at some point.
Great video. I've been trying -sref and --cref together, and they ought to get married! It's insane what can be done when used together. I've noticed photorealistic images lose details, though, when used with either. Older brother out of jail! Ha! Hopefully they'll make all this 'easier' by having sliders and such. And if you know anyone on the inside, PLEASE tell them to figure out the moderation. A half-naked, mutilated man turning into a zombie with flesh hanging out of his mouth in a bloody hallway is safe in the first generation. A woman in a modest bikini is not! And it was an MJ generated woman in a bikini. What do you mean? Of course that's what I tried first. [chugs Budweiser and burps]. Thanks for the video. 👍
Oh, I know-- the moderation is always going up and down! At one point I was trying to generate a new Pirate Danelle, and MJ was like: "Nope! Content Filter!"-- I mean, ok, she's cheesecake for sure, but c'mon now. MJ standard will generate stuff way more risqué than this! As always, though-- give it time. Launch day is always a little bumpy. I'm sure they'll figure it out. And yeah, I noticed that too on Sref/Cref-- but taking it through a creative upscale seems to help. That said, man it is powerful. The combo is really pretty amazing. The next hurdle is full pose control!
Very tempting to make a documentary about tech companies gleefully encouraging any level of gore, horror, and violence while brutally censoring something as mild as swimwear like their companies are run by 15th century nuns. Only the worst perverts love death and hate sex.
@@TheoreticallyMedia Much agreed, I implore more creatives to use the AI (actually software applications) or fall behind. I could make a comic or graphic novel in a few days or less at this point.
Kaiber 3.0 looks pretty good,. PIka can't compete with this. Pika is trash. What else have you got with video generation now that's at the level of Kaiber 3.0 or better?
Haha, if there's one thing I've learned by covering these tools, it is to never count any one of them down. They're all just one update away from blowing our minds. That said, yeah: I love that Kaiber looks at everything and says: "Y'know what? I'm gonna go down this other road..." Really appreciate that about them!
Super true! I’m going to talk about that in an upcoming video! Did you know when they were testing the THX sound system, they would play Empire Strikes Back in two theaters, one with THX and one without. The audience who saw the THX version reported the picture looked better. There was no difference in the film print! Pretty amazing.
I didn't have time to try it, but I saw some promising results where a few were trying out the "Sora Blue Alien" which, admittedly, has at least humanistic qualities to it. I think Fantasy Characters should work? As long as you don't stray too far off with multiple limbs or-- y'know, oddly: A Cyclops. Have tried many times: Midjourney hates those one-eyed giants!
Dragonborn are humanoid dragons/lizards, so hopefully that will work. On the multiple limbs note, I would LOVE to have a way to do that well. I have been trying to make a good displacer beast for months and it is so difficult without splicing images together in Photoshop...@@TheoreticallyMedia
agreed-- they've been talking about a ControlNet type pose styler in the near future. That's a pretty big deal. I'll admit, I'm not the hugest fan of the newer prompt style. I get why they did it, but I felt like there was more control in the older style. That said, --sref goes a pretty long way. I think it's almost time for a full MJ refresh tutorial...
Hey, so I'm new to Midjourney and as I was following along with your tutorial I noticed that when I go to the website I don't see the create tab that you have. What I see in the top left corner is only Explore, My Images and Rate my Images in that order. Would you have any idea why?
So currently you need more than 1000 images generated in MJ to use the website, that SHOULD change this week. It’s been in Beta for a bit now, and I believe the doors are opening for everyone shortly. Sorry you ran into this. The timing kinda sucks, considering we were only a week or so away, I had to choose to go with the web version or the Discord version. These videos often get referenced for months and months after I release them, so it just made sense to go with the web version, since that will obviously be the primary platform pretty soon. If you have done more than 1000 images, try alpha.midjourney.com
Hi TIm How are you today? Now with consistent characters and consistent style in Midjourney Is there a way to make 30 consistent frames to make a shot that makes a video when combined? thank you.
Hey! I'm good! So...probably not, only because there is no way to puppet pose a character across that many generations. Best bet is still to take the image to Gen-2, Pika, or even Haiper (did you see that video? It's FREE! ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-nU7lCnVmcBI.html)
Coming at some point! I’m excited about it too, since the claim is that it’s better than any other one out there. (They did say that before Sora, though!)
Yeah, they're still waffling between the two, but the site is almost 100% public. The Discord thing I think, for MJ, was really pivotal for them early on. It's the thing that built their audience and core fan base through the community aspect of Discord. They've long since outgrown it and I'm happy to see the website finally coming on line!
It sort of works with real photos, but tends to scramble the faces a bit. You can check out the part with my fictional older brother towards the end of the Midjourney section to see the results. It's weird though, I'm not sure how they're detecting a "real" face versus a synthetic one. However they're doing it, I'm impressed! I'm going to see if I can fake it out by taking an image of me, faceswapping onto a MJ image, and then --cref'ing from there!
You said that it does not really work for photos of real people (e.g. you and your older brother example). How is this different from uploading any Midjourney or other AI generated character reference image/s? It can't know and does the same processing regardless of where the image came from, or?. Perhaps I misunderstood something. Thanks for the great post 🔥
I think -sref is the closest thing yet to copyright infringement, actually allowing this feature is akin to music sampling. The music industry requires royalties if you do that and RU-vid will copyright strike channels for that also. The visual arts industry will surely react to taking their member’s visual work as a direct input into AI software. I think currently it’s the Wild West but heavy restrictions or licensing fees will certainly emerge.
My experiment with Cref did not go to plan. I had a picture of a woman taking a selfie on a beach and used that character as a female cyberpunk on the roof of a skyscraper. All the women were different in the cref examples and they were all in bikinis and straw hats. Not very cyberpunk. I will have another try tomorrow when I get chance.
Damn! Midjourney Character consistency looks great and interesting, it just a shame that they are so expensive! I guess when you are one of the best product out there you can charge whatever you want...
Yeah, I hear you. They have a line to straddle and I’ll admit, sometimes they get heavy handed. But for the most part, I think they usually course correct when they get it really wrong.
gonna dig into multiple characters in a follow up video, but for now: Yes, it can be done (sorta)-- just start with a 1:1 image and pan/extend and add another character in via inpainting. You can also use --cref for that!
Currently you need to have generated 1000 images in Discord, since the site is still technically in “beta” but it is supposed to open up to everyone very soon. I know, it has been a little weird doing MJ videos, since we’re in this transition period. But, I figured to future proof it a bit by using the website. If you have generated more than 1000 images, go to Alpha.midjourney.com to sign in.
So I need you to up date this video a bit, in order for people to get consistency with the clothing they need to negative prompt with the word -- no fashion . It took me all night trying to find a way to figure it out! Because you can't prompt clothing with a negative.
Yeah, all of them get weird about accurate weapons. A lot of times you just end up with Sci-Fi rifles as well. IT does get a little annoying when you're trying to do historical images and some Revolution-Era guy has a Laser Rifle, haha!
Well this was a long expected party!!! Curious, has anyone seen anything recently on using AI to enhance CGI or video-not just the resolution? It was looking promising mid last year but I haven't seen much since.
That's kind of something I'm experimenting with in Kaiber 3.0. The "Transform" feature has something like that, so I'm going to take it for a spin using some really old school/bad CGI cartoons...curious to see how it comes out! Keep an eye out for that video coming up soon!
It's astounding if you think about it that a whole year later corridor crews method for video to animation still seems to be the best if you need professional quality output?@@TheoreticallyMedia
Appreciate The tip regarding Inpainting the Head, will try that now. .. As the first issue I ran into with CREF is, it doesn't support two people in the Image. Like a Man and Woman. It turned them into Brother and Sister which was creepy since Im working on Romance Stories lol
Haha, apparently the hack there is to use a Pan/Extend and then add your second character in via inpainting. I'm not sure it FULLY works right now, but it is at least a workaround for now! (It's MJ, there is ALWAYS a workaround, haha)
@@TheoreticallyMedia So far the inpainting the 2nd person hack is working, unreal!, however its very hit or miss as Midjourney keep spitting out random error messages when trying to render the inpaint. Weird because it worked flawlessly for about 15 to 20 minutes just a little while ago
If you come back to this subject it would be a great help if you could show this workaround in both discord and the website. I have not got it to work yet. Having just one character in a scene is not going to hack it for storytelling
It can, particularly through training LoRAs-- A different method, obviously, but I like to think of them as different tools in the toolbox. MJ's cref feature is really interesting though, considering it takes zero time to train. I'd be curious to see how they're pulling this off...Although, MJ being MJ, I doubt we'll ever find out!
1000%! And totally-- MJ really nails it with the ease of use! That whole thing of just dragging the character image over into the prompt bar? That's such a small thing, but also such a massive time saver!
Heya Louis!! Have you played with the --cref feature yet? And, I'll be curious to see your thoughts on the upcoming Kaiber 3.0 dive. I almost think this is a perfect spot for AI Video-- where you can really lean into the surreal!
@TheoreticallyMedia I've not experimented with it yet. I generally don't have a need for consistent characters but in the event I ever would, it's good to know there's a tool in MJ that can help to make that task a little easier
Oh wow-- really appreciate that! I was saying in another comment that Midjourney videos always feel like "home" to me! Whenever I get the chance to cover a new feature, I always have the most fun making them! I really enjoy all the tools I cover these days, but I'll always have a special bond with MJ!
Now these are game changers! I'm making a music video for our band but I know we're on the cusp of some really cool stuff. I think this rounds everything out nicely for a real run at it now.
a lot of the video editors have them installed as effects-- both Premiere and Davinci. I think Topaz has one as well, but that one is super costly. I'll look around for a platform based solution as well!
Oh, thank you so much! I wanted to do a full worksheet with it, but kind of ran out of time. I'd still be working on it now if I did! That said, seriously appreciate the comment! It really means a lot!
@@TheoreticallyMedia I REALLY enjoy your worksheets, so much so I always happily drop some $ for them. I use the negative prompts from your Pika Worksheet all the time.
Yoooo thats sick thanks for the Kaiber update I have a TON of credits sitting on the wait for this exact update. Really really nice improvements there for them ! Edit: yeah that flicker is very very low compared to what they had before it should be easy to use comfyui to interpolate and enhance that out im super impressed with this update
For sure!! Make sure you ping me/yell at me about it! (And keep with it)-- I'm constantly struggling to keep up with messages, so seriously: Don't stress about hammering me! (my wife learned that lesson decades ago! HA!)
Wow! There is a whole science how to use those tools for your purposes! It reinforces the idea that cinema/photo industry will change for ever! You are doing great service to people by explaining how it works! Thanks, Man!
Appreciate that! And for sure, this is such a wild time for anyone in creative! I think there is such an opportunity here to experiment with these tools early in their development to really help define what the vocabulary of using them is going to look like in the future!
The fact that the consistent characters process is now streamlined is so great, all though I do fear some development within ai (scammers, ai generated revenge corn, ai generated defamatory attempts), I am very excited about the creative aspect of it, the dream of making my own movies is becoming more and more possible by the minute. As always thank you for the video! I look forward to the Kaiber video coming up! :)
