CONSISTENT Characters Still Work in Midjourney V5!

preview_player
Показать описание
Folder: 2023-03-21 – Consistent Character in v5

🌐 Check out the full blog post:



🎬 Character Design Series 🎬

📺 Recommended Related Videos & Playlists 📺



So you started playing around with version 5 of Midjourney, but for some reason, none of your old v4 prompts seem to work the way you want them to.

And worst of all, you just can't seem to get consistent characters anymore.

Don't worry, we're all in the same boat here.

Version 5 is clearly putting many of us to the test.

Version 4 did so much of the heavy lifting for us.

But Version 5 forces us to grow and improve our skills and that's a good thing!

I was curious to see whether any of my previous methods for consistent characters in Midjourney v5 still work.

I also wanted to see whether I could bring Carla Caruso back to life.

So if you're curious to see what worked and what didn't, then stick around and watch this video.



⏰ Timestamps ⏰
00:00 Bringing Back Carla Caruso
00:45 Character Styles in v5
05:43 Consistent Characters in v5
07:46 Controlling the Character in v5
08:53 Creating Portrait Scenes in v5



#midjourney #midjourneyai #midjourneyart #aiart
Рекомендации по теме
Комментарии
Автор

Hey Christian, thanks for all your work.

A few notes: I'm working with realistic models, not illustrations, so I"m not sure it's the same, but I suspect it is, so here goes:

If you want a very consistent character, 4 URLs isn't nearly enough. I'm using up to 16 URLs...the characters, well, they look like the same person in a photograph.

If you want full body images, it's helpful to find a photograph of a full body image of a celebrity or whatever, use a cutout if you can find one, or remove the background yourself. Then add that URL to the prompt. If you search online for images, most photographs don't include the feet, so the references MJ has in its data base tends to have the character or person either as a portrait or cut off at the waist. Using a full body image in the prompting gives it a reference to use.

Once you have a very consistent character, i.e. all four images generated look basically identical, THEN you can add to the word prompt things like "put her in Paris" or "on a shooting range with a Glock" or whatever. MJ still messes up the way a character holds a gun or any object -- -But the character will be identical and you can reroll a few times and eventually get a good one. I'd say one in 8 or 12 have my homicide detectives holding a glock reasonably well, but regardless, the characters are the same.

I pulled some of the images I liked (the URLs) from MJ v 4 and added them to photographs of celebrities I liked, mixing a somewhat cartoon/illustration with a real photograph can then be massaged towards the style you like, just remember to keep the character consistent, you need these 12-16 URLs in EVERY prompt....then MJ is forced to create a composite image and, viola, they end up being the same person or character.

Of course, like everyone, I'm still learning and I'm sure in two months, with V 6, we'll have to relearn a bunch of things...haha...

OH. one more thing, instead of using "Marvel Comics" you'd probably get a much more consistent style if you tell MJ to adopt the style of a particular illustrator (or character) as "Marvel" is too wide, or at least, that's what it seems to me. As I said, I'm working more with things that require a photorealistic model (so I was so happy with v 5 and had to redo a bunch of stuff) but I suspect that if you tell it to use the style of <insert name> it'll get closer to what you're looking for.

Well, thanks again, I"ve been watching your channel with every new video and you've helped me so much.

michaelsbeverly
Автор

We love the current MidJourney V5 Pro version with all of it fine adjustments within prompts but when the non pro version becomes the default I have been told we can still access the Pro version. Keep creating your changing the world at the speed of light.

GoapprStudios
Автор

The v5 also doesn't understand wat is "boy" and "girl" and if you not sayed directly that you want male or female character it will blend and mix it into image in differetnt proportions some times to the point when you just want to "unsee this". This version of a product making us use more words in description, that's for sure.

MayStrongSky
Автор

Great video series. I have been using Runway AI in an attempt to create reliable AI models but hit a wall and started playing with MidJourney. Your videos have been a big help trying to navigate this other style of AI. Still running into walls but the walls get pushed back a little more each time.

ryguystuff
Автор

It took me a while to find out the importance of the order of the prompts. "scifi fantasy armor" will give you quite a different result to "fantasy scifi armor".

chrisanderson
Автор

@10:21 the image on the right truly does look straight from a high quality comic book. You can visibly see single hatch line strokes to help build form, a nice line weight outline and nice lighting/shading overall.

How can you keep that style consistent across all the images? Any suggestions or thoughts?

Rayartz
Автор

Are you creating all new prompts every-time you change something or are you remixing it and adjust the prompt?

natalieyetinson
Автор

Thanks. I noticed you struggled with getting Midjourney to draw Carla in her full body pose, I wonder if there are any tricks to force MJ to show the full body, perhaps pose style names

icedanilzation
Автор

At 10:27 I think you made a mistake. You say you have changed the aspect ration to "--iw 9:16". Don't you mean "--ar 9:16" ?

SpookiePower
Автор

I'm about to watch this video, hoping it will help. For some reason, MJ is ignoring my seed instructions and I get lovely images but with a different face. Even if I specify the seed number, the images turn out with a different seed.

bambubombon
Автор

As I understand it, the fact that Migjornei created generally different characters with the same hairstyle does not bother Crisitana. What if we have something much more complicated? bad leadership

АнтонСтепура-рд
Автор

I'm an AI engineer, where can I sell my extraordinary prompts? I tried promptbase, but it's not available in Thailand((

KatLoree
Автор

I cant get the image based on a photo of mine ... how to have consistent character based on my personal photo ?

aibronman
Автор

Ugh stop with the consistency videos.. move on. I like your channel but tired of these

GrisslesWorld
Автор

I watched it 2 times, it was very useful

CrackleJue
Автор

Agawam Skyline done with Midjourney 5😊

jamesmonahan
Автор

Brilliant, thank you Christian. Been struggling with my characters in V5 until you dropped this. Thanks for being such a damn great MJ guru 👊

GrimmHollywood
Автор

This has been a really interesting experiment. I've been trying it out with a character that I got from MJ earlier. One thing I've noticed is that you really have to battle Midjourney to stop it making your character younger and more attractive with every iteration - I'm trying to make a somewhat normal looking guy, and the bot really wants to turn him into a supermodel 😂. I've been putting in age prompts (E.g. "25-year old man") and even then, I'm noticing you have to age the prompt up by a few years as it still wants to make the character younger.

orirune
Автор

Ah, another thing about POC. In creating a consistent white character, it seemed to me that it took less images -- why, not sure.

When I needed to build a Black male and an Asian female, both as homicide detectives, it took many more images to get the character consistent. BUT, once done, once I had a prompt with like 20 URLs, the images came out consistent, pretty much every roll. Then, adding, "in dress blues" or "at her desk on a computer" or "with a glock on the shooting range" all worked AND kept the detective as the same person. The minor variations are all what we'd expect with a real photo session with a live model, sure, some variations depending on lighting, mood, etc., but recogniable the same person.

michaelsbeverly
Автор

So the rules have changed. What do you recommend using V5 - is Parts 1-5 still working prompt wise or is it so different that I'd have to do it totally different? Thanks for explaining all the -- additions (--ar and so on)

TheTimedCarbGuy