Midjourney v5 -- Prompt Tips For The New Version! (Things Have Changed!)
Вставка
- Опубліковано 15 бер 2023
- Today I've got some Midjourney V5 Prompt tips for you! Prompting is a little bit different in V5, and in my opinion has changed for the better, allowing for a more conversational approach to using Midjourney. So let's dive in and see the latest on image prompting in Midjourney, PLUS take a look at how Aspect Ratios have changed...and when I say changed, I mean: we have ALL OF THEM!
Check out Brandon Woelfel here: / brandonwoelfel
And here: shopcatalog.com/people/brando...
Follow Me on Twitter: / theomediaai
AFFILATE LINKS:
Camera: amzn.to/3yXMDY2
Microphone: amzn.to/3K1jIZm
Audio Interface: amzn.to/3lDX9kf
Coffee: amzn.to/3JZuBeq
Rounding out, I did use some background music in this video. It's actually music I composed, and if you'd like to hear more, you can check it out at the links below, and if you ever want to use the track for your own stuff, just shoot me a message and you can have them for free.
SPOTIFY:
open.spotify.com/artist/5Q9MN...
APPLE MUSIC:
/ citizentim
-------------------------------------------------
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!
-------------------------------------------------
► Business Contact: theoreticallymedia@gmail.com - Наука та технологія
What do you think of V.5 so far? Also, my hair was very sad today...😆
You talked about that you had a promting server? I cant find it. Btw! A very informative video :D
Ah, I think I'm referring to my Cinematic Prompt Formula video-- that's here: ua-cam.com/video/doeMSB96JOA/v-deo.html. And THANK YOU!!! 😀
@@TheoreticallyMedia Oh yeah. Thank you so much 😁👍
Impressive
Tim :) Thanks so much for creating this video. I really appreciate the info you "bring to the table", really useful and insightful.
Thanks so much Graeme! And I really appreciate you watching! I'm looking forward to doing a lot more digging into v5!!
You are the only person I have found that can succinctly give useful info. SO glad I found you.
I need to screenshot this comment, blow it up to poster size and hang it in the living room, so I can point to it every time my wife asks “what are you blathering about?”
Haha! Thank you so much, really!!
Exactly what I was looking for! Been having difficulty getting consistent results in v5, and this really clarified every question I had! Thank you so much!!
That's fantastic to hear!! Thanks so much for that comment-- and please, if any other questions come up, don't hesitate to leave a comment! Always happy to dig in and find the answer! My thought being: If you have a question, someone else certainly has the same one! Love to help our little community!
Thanks for keeping us updated on new features!
Thank you for the watch!! Gonna keep a close eye on all the new improvements on deck for v5! My guess is that I'll wrap up just as v6 hits!! haha
@4:30 - for a more retro look to the imagery try these prompts: real 1960s photo, real 1970s photo, etc etc or even types of film grade like Polaroid photo and different camera settings like: f/1.8 (for a shallow depth of field,) Kodak Professional Portra 400NC (for more of a 35mm film look). On another note, v5 doesn't produce image seeds for upscaled images at the moment (by commenting you images with an envelope emoji), however, it does work with the first render of grouped images (the 2x2 grid)
Thanks Tim. I appreciate your video. It gave me good info to use.
Thank you, Robb! Glad it was helpful! Hoping to have a new one up tomorrow!
Thank you, brother1 great video and info
Thanks so much for checking it out! And stoked that it was helpful!! Cheers!
What you said at 1:06 -- I had been looking for that (if you could just toggle between the 2) -- THANK YOU!!!!
Oh, EXCELLENT to hear!!
Nice content, mate. Thanks!
Happy to hear!
Great insights, thanks
Thank you for watching and dropping a comment!
Love this thanks 🙏
Missed this comment ages ago, but just wanted to say THANK YOU!
Love your tips and tricks, but I also love your bass and guitar in the background.
Oh, awesome! The guitar has a funny story, it’s an old 80’s Peavey, swamp ash and heavy as hell. I got it off FB marketplace from this old timer for $200. As I was leaving he said “you know that guitar was owned by The Nitty Gritty Dirt Band.”
Seems like such a weird thing to lie about, and he already got paid, so…I guess it’s true? And while I wouldn’t call that band “famous” it is weird to think I own a guitar that has likely travelled the world and been on records. Haha.
Been meaning to get back into doing some AI music content, so I’ll try to break it out for you soon!
@@TheoreticallyMedia Nothing wrong with Nitty Gritty Dirt band. I was exposed to them when I was a kid because my dad, also a guitar player, liked them.
It's really fantastic. I will try with this Blue Willow and lets see the result. Recently I used BW for making images and I am amazed by the results
I still need to get back to BlueWillow for some tests. I tried it out when it first launched, but that was obviously pretty early in. I know they've done some pretty big updates, so I'm long overdue to circle back and try it out again!
I'm new here. Great video, straight to the point. 👍
Welcome aboard! And you aren't new, you're just in time! Haha, I do apologize, the latest video is like 13 minutes long-- but I didn't fill it with a lot of waffling. It needed a lot of info dumped, so the runtime went a little over!
Good stuff, subscribed.🌟👏👍
Thank you very much! Happy you're here!
The images are so realistic that I still don't believe 100% they are not real pictures in some cases.
I know, right? Have you run across the thing where MJ produces a person that kind of looks like someone you actually know? That’s super weird!
You mentioned rule of thirds, but is there a way to tell midjourney to actually put a figure or portrait on the left (or right) side with doing a 16:9. I always end it with it being basically in the middle of the image.
This is awesome, thank you! How would I put an image of myself into an image I wanted to create?
So, Midjourney is not great at copying faces from reference photos. I think it CAN do it (as proven by images of famous people), but I think as a safety measure it scrambles facial features when you upload an image, so no one can go crazy with deep fakes.
So, depending on what you were thinking of, you might want to look into Leonardo's Image to Image (ua-cam.com/video/uKz7qtEINs8/v-deo.html), or if you wanted maximum control/quality, you could generate a character in Midjourney and then Photoshop your face in.
I am going to do a video on a technique called Photobashing next week-- so I'll try to incorporate some photos of myself to see how the results come out!
Any tips for getting different aspect ratios, typed in the prompt different aspect ratios and seems to only output the 4 different 1x1 square photos. currently using v5 with the same settings you recommended.
That's odd. If you put in --ar 16:9 at the end of your prompt (dash, dash, ar space 16:9) you should get a widescreen look. Let me know if you keep running into trouble!
Awesome video. Thanks for sharing. Where can I find the prompt channel?
Awesome! Glad you liked it! Prompt Chat Channel is on the right hand side of Discord-- should be under the "Chat" Channel tab. It should go: Discussions, Philosophy, and then Prompt Chat -- Let me know if you have trouble finding it!
Excellent video, man! I'm new to Midjourney this week but as a writer I haven't really struggled to get some fantastic results. Two areas I'm struggling with, however, are negative prompting and aspect ratio. It doesn't seem to consider either of those parameters despite correctly formatting them and including descriptive language to support the parameters I include on the back end. Are there any secrets or even common knowledge work arounds that a newbie like me should know?
Writer here as well, so I might be able to help! So, one quick tip is to not write in prose, or honestly, even grammatical correct!
The larger your prompt, the more apt it is to ignore words. So a prompt like “the view from a barn window, as storm clouds gather in the distance” should be minimized to “POV, barn window, storm clouds in distance”
By minimizing the prompt, you’re more apt to have it respect (dash dash)no commands.
Aspect Ratio, to be honest, I’m not sure why you’re running into an issue. (Dash Dash) (space) (first number) (colon) (second number) should work. Let me know if you keep running into trouble there!
@@TheoreticallyMedia I do find that a lot of the times simpler prompts will yield the highest fidelity but still struggling with my AR frequently, strangely enough. It's squeezed a few out in the right dimensions but no regularity. Seeing a lot of different ARs from others so I know it's just user inexperience at this point..
@@foxross4468 What AR are you aiming for? I have seen that some weird ones, like 4:1 don't work as well-- but that's also a pretty funky Aspect Ratio, so I can understand!
@@TheoreticallyMedia I had been attempting 16:9 and 7:4. Now trying 2:1 based on your stuff, but so far that's only hit successfully once. Your cinematic prompts video has, however, has helped me produce more aesthetically accurate images though--so thanks!
Nice subscribed 🎉
Thanks so much! I have some really fun videos coming up! Glad to have you here! Welcome!
V5 seems slower because it's upscaling all images in the background before they show the thumbnails. AI technologies like Blue Willow will become more and more popular over time. Blue Willow recently launched V2 with a brand new model upgrade. Now I got more better-quality images with more aesthetic outputs
Weird, it seems quite a bit faster for me than V4
Any tips for general image prompting? I feel like v4 did image prompting better maybe I’m doing something wrong but I still prefer this overall model
Generally I've found the mostly standard stuff: Words at the start of the prompt are treated with greater value, so I'll generally start with the style there, ie: Cinematic Still, Illustration, etc-- then I'll follow that with a character description and the action they are taking, followed by camera and lighting details. I have a video on how I treat prompts for Cinematic Images here: ua-cam.com/video/doeMSB96JOA/v-deo.html
But in general, I kind of use the same prompt formula for any other style I'm working toward. Just replacing the medium's nuances in the various sections. Let me know if that makes sense to you!
@@TheoreticallyMedia thank you!
Now we just need consistent photo real characters. Had an idea to do a cinematic type comic book but cannot get consistent characters unless they are celebs.
Yup. That and posing is always tough in MJ. I actually used to do indie comics, so that’s been a long running dream of mine as well! I’m working on some tests for a tutorial on comics, but honestly I’m leaning towards Leonardo as the tool to accomplish it. MJ is too inconsistent where it comes to storytelling.
@@TheoreticallyMedia Leonardo ai looks pretty solid. I'd do a subscription plan if it was like Midjourney. I feel like I would burn through 25,000 tokens before the end of the month. I like the relaxed mode in MJ. Pretty much unlimited generations for $30 a month.
Does MidJourney block/ignore requests for images of Grogu?
So weirdly, yes-- and no. Image prompt for Grogu got me a knock off from a Dreamworks movie, but a prompt for "Baby Yoda" got me Grogu. I guess Midge is only a casual Star Wars fan. haha.
Truthfully, I think the name Grogu just hasn't been attached/tagged yet. I think that means we can point the training data to pre-Season 2!
@@TheoreticallyMedia Thanks for the investigative work. 🙂
Yes it makes sense that there isn't enough data relating to Grogu by that name.
I'm tempted to get a subscription, but from what I can see, V5 doesn't have relaxed mode. So you'd be using an awful lot of your time since it renders natively at high resolution.
Not to worry: V5 has a relax mode! I use it all the time! I will say that v5 is a little slower in relax than v4 was, but that also might be that there is more strain on the system as every MJ user was playing around with v5 in the opening days.
You can use relaxed but only at $20 sub or higher
How to stop a previous prompt from appearing at the end of a new prompt? Help, please!
I'm sorry, can you explain a little more in detail? You can't just delete the previous prompt? Generally, every time I prompt, it is with a blank slate. I might be able to help if you can clarify a bit!
@@TheoreticallyMedia Hi! Thanks for your immediate reply. Each time I prompt, it is a blank slate. However, after I type a new prompt and click enter, a previous prompt appears at the end of the new prompt.
@@gregorymiles6457 That IS strange. So if you type in /imagine a tree in the woods and run it, the next time you prompt it'll be:
/imagine John Wick, holding a machine gun, a tree in the woods
Is that right? Because, that's super weird.
@@TheoreticallyMedia Yes, this is exactly how my prompts displayed in the queue.
Can Midjourney generate vector image?
It can. Try starting your prompt with Vector Illustration, or Flat Design, and that should get you there.
@@TheoreticallyMedia Thank you, I'll try that, will buy a subscription plan, I think.
one more thing, what about 3D models? Don't get me wrong, I am not greedy, it is astonishing what is does now, I am just curious where it is going, if Midjourney is going the way, it will do animations some time, maybe in V7 , or make us a complete movie maybe in version 17 ? Or are there other products more specialized in these?
@@nick066hu So Midjourney has stated a few times that they're looking to be the "Apple of AI Images"-- To that, I think they're aiming to maintain quality image output and do so in a simple/easy manner-- so, I don't think 3D is in their plans. That said, Leonardo.AI did just announce that they are going to have a feature to skin 3D models. Like, you can't build it in Leonardo, but you can paint/texture them there.
I do think that we're headed in the "Full AI movie" direction, but it is more that there are a bunch of different products doing different things. At some point in the future, someone is going to bring the whole thing together though. It's funny you mention that, since that's going to be the topic for Thursday's video-- unless, something super major happens in AI in the next 48 hours...which, given the speed of AI, is super possible!
Im experimenting a big problem, i use this prompt, i also change the order, but it always give me a room with David Lynch inside, and no phone around, what happened to midjourney 😅**cinematic still, film by David Lynch, white male stand, in the living room, talking with a smartphone, wide angle, style dramatic, 35mm, 8k --ar 16:9 --v 5
I think part of the problem is that you don't have a name for your "white male" so MJ is saying "Oh, I see David Lynch here!" and filling him in. I ran your prompt and got similar results to you. Just playing around for a second (sorry, not a lot of time right now) I got some results close to what I think you're looking for with this:
"cinematic still, film by David Lynch, Young Bill Pullman, in a living room talking on iphone, wide angle, dramatic lighting, 35mm, 8k --ar 16:9" -- Still not 100%, since I'm getting OLD Bill Pullman, and only 2 of the images he's on a phone, but it's a starting point!
@@TheoreticallyMedia thanks, maybe i can try with "generic male", and maybe add some traits, it s working if i put a woman. I also try Simple text like "a woman talking at the phone" and it give me shutterstock like images, perfect, but if i use the same phrase and add director, living room, it don t work anymore, it s like the first weeks midjourney dont had this problem, i m a bit confused 😂
@@Diego-de9eu sorry, on mobile so I can’t try anything out right now, but an idea- I just posted up a video on image ref/weights, check that out and it might help.
The idea would be to take your “shutterstock” output and then run that as a weighted image with your prompt. See if that works?
I m not sure, i wrote "John wick night club scene" and i had the same result
Maybe we should try some weirder alternates then? Like John Wick, on a spaceship, with a battle axe" Haha, see what we get then!
Slightly off topic, but I am now super pumped to see John Wick 4! That dog better live!
The text ou type is very hhard to read because of the background. please make the bakground simple whenever you are using text. Thank you for all the knowledge.
Excellent to hear! That one is a bit of an older video. I've gotten the text pretty clean at this point! Mostly because of comments like yours, so honestly: Thank you very much for the feedback!!
I just love it when there is a screenful of text, and the narrator is reading it, but then the screen changes before the narrator (or we) can read to the end. If it's not important to your message, don't put it on the screen.
Did I do that? Do you mean the image prompts?
I cannot confirm this. What bothers me about Midjourney is that many instructions are simply ignored. It ignores sustainably that I want to have a full body picture and not only a portrait from the hip ... The addition of "in the style of" is also completely ignored. While even Kurosawa and Tarantino should differ, at the latest Frank Miller should look completely different ... The result is almost identical for all "in the style of" additions ...
The Promt was: "A gloomy, mystical and already somewhat threatening forest. A ten-year-old Viking girl is standing in a clearing where fog is wafting. We see her whole body from head to toe. She has torn, dirty clothes on and wears no shoes. She threatens with a sword in her hand and screams a loud war cry. She looks menacing. Her long blond hair is braided. In the style of [one of the mentioned guys] --aspect 16:9" (V5 is set by default and therefore not mentioned right now in the prompt above. The Bot did include --v 5 ...)
I don't disagree with that-- Although, I think there are some ways to get the shot you're looking for. Your comment inspired me to take a swing at it, and now I'm just making.a whole video about it (haha)-- I'll swing by and reply when its up!
@@TheoreticallyMedia I have already been able to isolate the problem somewhat. For one thing, Midjourney doesn't seem to be able to create full-body portraits in landscape format. If you changed the aspect ratio to square or portrait, MJ had less problems with the task.
And also, the desire for a "battle cry" causes the camera to move closer to the subject than if the desire is omitted.
Still, I find it amazing how persistently MJ simply defies some commands ...
@@erikschlicksbier I also think it's a matter of prompt order as well-- I'm actually using your prompt as an example, but I did futz around and change the order around to see the results. And then ultimately, abandoned it for something else (haha, sorry!) and I think I'm dialing in. Some REAL problems with Frank Miller (I'm a big comic nerd, so as soon as you said FM, I had to get to work!)-- but moving to Mignola created some real gems.
The Battle cry...yeah, that's tough one as well! MIGHT have a solution for you!
@@TheoreticallyMedia I wasn’t set on Frank Miller anyway. I just tried him for his destinct recognizable visuals after Park Chan-wook, Tarantino and Kurosawa gave the exact same visual style. My guess is that maybe the Miller-style is excluded due to copyright issues?
Plus: the viking girl seems to include a visual style in itself by Mj. When I tried to apply the style of the instagram guy you mentioned in your video, it changed not that much. Not that much like for the 70s guy at least (which I could reproduce quite like you had it).
"He died, so it's okay to use his style, it's the kinda thing that gives Midjourney it's look".
I've noticed the paranoia kick in...
Ha! Nah, I met Darwyn a bunch of times. Even drank with him and watched him get in a bar fight once. He was a real old school analog artist.
I have no doubt he would have hated this.
If there was anything you saw, it was a little sadness at his passing. The guy was a hell of a writer and an artist.
Edit: stupid spell check.
@@TheoreticallyMedia 👍 bless 🙏
midjourney doesnt seem to like guns... well its obvious it doesnt like violent images but i guess it relates holding firearms to violence and (imo) intentionally messes it up whenever you ask it generate someone holding a firearm :\
I've been hit or miss on it in the past as well. Awhile back I did a fictional documentary about a film called The Gunslinger (based on the Stephen King novel), and obviously being a Western, there were a lot of guns in it.
The guns were a problem here and there-- particularly guns shooting.
The other thing I've noticed is that guns are generally fictionalized in MJ. You'll rarely see something that looks like a real life firearm.
But, y'know-- I get it from their perspective as well.
@@TheoreticallyMedia i tried making rainbow six siege images and i was able to generate firearms that stood alone but when i tried having a character hold them it was a completely different story smh oh well i guess
Nobody has spent a lot of time on midjourney, it came out last year. Why do people talk about it like they studied it at university.
Ha! Well, to be fair, people teach new and emerging concepts all the time. The trick is to relate it to something you do have an area of expertise in. Also, while MJ is new, computer generated imagery has been around for quite awhile. The method and techniques might be novel, the underlying concepts aren’t.
Can it generate erotic photos ?
Depending on your definition. I mean, some people are really into feet. 😀 No judgement! But, if you're talking about non-clothed, then no.
this thing is like tik tok. imaginary dopamine, feeling like you've done something when in reality you've done nothing. people will call themselves artists for things this program generates and really think it was them.
Eh, half and half. I agree that if you're using AI image generators JUST to create images, than sure-- but at the same time, it is cool for non-artists to be able to invoke something from their imagination. There are also others who use these tools to create larger works. Like, creating images for music visualizers, or video presentations. In that regard, I almost look at it like sampling-- taking something that was, and recontextualizing it into something new.