Create MULTIPLE Character Scenes - Midjourney Character Design

[@adriannairda7922]

Your videos set a new highest level on Midjourney prompt engineering. Really inspiring, I'm able to generate character scenes almost perfect with the skill you taught, thank you.

[@TokenizedAI]

You have no idea how much comments like these mean to me 🙏🏻😊

[@crobinso2010]

They need to allow Object creation, so that one character is created independently from the background or another character, which you can tag as other objects. After creating objects without bleed, then you can prompt per Object or "hold" an object while changing the others. Object creation could be on the Pro version, for example and be a huge incentive to buy it.

[@TokenizedAI]

I agree, though I'm not sure that currently possible with latent diffusion models. Inpainting and Outpainting would already be a huge step forward.

[@danke1150]

You can use :: as a separator

[@saintshing]

Can we first generate a set of concept arts for a character and then use dreambooth/textual inversion to add the character to midjourney/stable diffusion by training with those concept arts?

[@TroyHardingLit]

The concept of mirrored multiprompts is truly inspired. I tip my hat to you.

[@TokenizedAI]

Ooooohhhh...I think you just gave that method a beatiful name: "mirrored multiprompts" Love it! 😃

[@huibert_devet]

short explanation? 😉 Or in what your video I find this?

[@TroyHardingLit]

@@TokenizedAI Go ahead and take it if ye like it. :) *thumbs up*

[@howiedewitt_]

Wouldn't adding a weight of say ::2 accomplish the same thing as including the same prompt twice?

[@TokenizedAI]

@@howiedewitt_ It's not the same prompt. It's a different prompt, using different words and sentence structure to describe the same thing. The whole point is to use multiple different descriptions, increasing the likelihood of getting an image that meets your expectations. If you just use 1 segment and your text prompt produces something very different than you expected, then repeating the same segment over and over again or giving it more weight is not going to solve your problem. It'll only make it worse.

[@max-imize]

Love what you can get out of MidJourney! 🔥 Prompt King!

[@TokenizedAI]

Thanks for the feedback 🙂 Congrats on the great channel growth btw! 🚀

[@max-imize]

@@TokenizedAI thanks a lot! 🤖

[@SiraEkabutPlus]

Totally agreed!

[@heretolearn4214]

The Goats !!!

[@StygianStyle]

I've tried stable diffusion installed on my machine with a few of the best looking models. Midjourney still manages to make images that are more expressive, detailed and beautiful. It seems like it's done by a more talented artist than what I can do with stable diffusion.

[@TokenizedAI]

The power of MJ lies in the proprietary embeddings tha they use.

[@giri.goyo_yt]

This is why I'm a devoted follower of Tokenized AI, and this is EXACTLY why I started my Midjourney and this is going help me finish my book in a huge way. THANK YOU so very much for all your work!

[@TokenizedAI]

So happy to hear that these videos are helping you make progress 🙂

[@CorporalDavis]

Way to go...going to try this as soon as I can.

[@Leonardo-13]

It's not just 3 days that were hair pulling.... but also the whole effort to produce the video, so that we understand it all better here. I have learned a lot. Thank you so much!

[@TokenizedAI]

Yes! Thank you! 🤗

[@corex72]

Awesome work man. I’m working on a story board. Arggggg

[@giuseppedaizzole7025]

Its not possible at today....great video! thanks

[@paris.granville]

"Mistakes are portals to discovery" I appreciate the inquiry spirit of your videos!

[@shoptimelessboutique5858]

OMG, I COULD HUG YOU RIGHT NOW, I AM ON MIDJOURNEY WOOT WOOT, I AM ABOUT TO CREATE A STORY CHILDREN BOOK WITH 3 CHARACTERS... SO I AM RESEARCHING- YOUR VIDEO IS ON POINT🤗🤗🤗🤗🤗. Thanks for sharing.

[@TokenizedAI]

Go for it!

[@whuwatwinwear6786]

I am new to the Midjourney world. Thank you for forging a path for the rest of us!!

[@Hyper5nic]

The first generated images are indeed marvel comic book related for the cover. The 'ugly' second images are similar to drawings on the inside of women oriented marvel comics. You could add 'cover image' or just 'cover' to get better results.

[@dccrowley]

Christian! wow. That is pushing the envelope. This is very difficult and you kept going. Thank you so much. One of the best videos I have seen to date

[@TokenizedAI]

Trying to push the envelope because most people just refuse to or they keep it to themselves.

[@loungevity]

Wow Christian , truly amazing, thanks for spending 3days working this out for us all, I shall now spend the next 3days putting in practice today's tutorial... Thank s again 🙏

[@TokenizedAI]

I hope you guys can all validate that this approach works more or less consistently when done right! 🙂

[@writerlass5924]

Genius! You are one step ahead of me every time and I’m so grateful you are leading the way and gradually cracking the MJ code!!!😀🙏

[@TokenizedAI]

Thanks! We'll get there. I'm sure of it 🙂

[@KruzSanchez]

Currently working on a comic book just for fun with midjourney and your videos have helped a TON!

[@TokenizedAI]

Glad to hear that 🙂

[@mahoneytomagao1006]

Thank you for all your hard work. This was such a very valuable information. I believe the answers are just around the corner and just like prompt engineers are breaking Chatgpt, Midjourney's secrets will soon be revealed.

[@TokenizedAI]

I sure hope so because MJ does make it pretty difficult. We're sooo close and yet, so far away from solving this 😅

[@psharp2468]

Here’s some more prompt magic for you to tinker with: - -S (then list a number between 0-100; this tells the ai what level of style freedom you like to allow it, with the “0” being the low end of the scale. - -C (then list a number between 0-100; this tells the ai what level of chaos freedom you’d like to allow it, with “0” being the low end of the scale. Both could help limit the amount of change that midjourney applies to a seed that you provide.

[@TokenizedAI]

I'm aware of --stylize and --chaos. I have found Stylize to be relatively useless for keeping things consistent because while lower values stick closer to the prompt, they are also less visually pleasing. And fyi, the scale for --stylize is 0-1000.

[@Mongloire]

You are hilarious with your persistence-and curiosity! Love your videos!

[@TokenizedAI]

Gotta keep pushing the limits. The value lies within the hard stuff.

[@undeziner]

loved this video, only yesterday i have learnt something from my 11 yo daughter. She wrote the prompt in a most simple, basic english language and the results were tremendous.

[@learnspanishwithdavid]

I have been watching tons of videos about creating consitent characters and they all explain exactly the same, then I tried myself and I got nothing, it was so frustrating and I was about to give up. This is the first video that explains the reality. Thanks a lot!

[@TokenizedAI]

Because it's easier to sell a dream and only show a result rather than actually reproduce and demonstrate it.

[@resortX]

I don't know, if the --iw prompt was available to this time, maybe working with images as reference, the image weight prompt can do it.

[@TokenizedAI]

It was but it wouldn't help in this context.

[@MrForestExplorer]

Respect to you for your channel. These videos must take quite a bit of work to make every few days. Well done. Some valuable insights. This is the advanced level imagery I am currently trying to make. Lots of experimenting. I've tried these methods... They work to some degree, maybe one good one eventually, after many re-rolls.

[@TokenizedAI]

It's definitely challenging and yes, the video do take quite a bit of time. But I enjoy the discovery process :)

[@camizanatta]

All I can say is... Thank you soooo much for all! A true master is not just the one who knows too much, but teaches his knowledge to others! (Sorry for my english, I speak a different language...)

[@TokenizedAI]

Thank you. I really appreciate comments like these 🙂

[@togus13]

Another dilemma I noticed is that if u say a red-haired boy and blond-haired girl in a prompt together. It always swip the hair color to each other. For some reason 95% of time is an issue

[@TokenizedAI]

Yep, the color-bleed and color-switching is a major problem.

[@CptDangernoodle]

@@TokenizedAI just finisywatching ar it occurs to me, what if for the three miltupromots, you use the first ine to exclusively describe one character in d tail, the secdon promot to describe the second character, and the thirs promot to mention both characters without describing them.

[@CptDangernoodle]

And sorry for my weird spelling, the youtube app bugged out and I cant see what Im typing, nor can I edit it 😂

[@melodytong7778]

That's awesome! A few days ago ,I put a cat and a rabbit together in a picture,it’s failed. Now I know the reason. Thank you so much !

[@TokenizedAI]

Glad it helped!😊

[@DeanWilson]

oh wow, i just asked this question about an hour ago in your other video. Awesome

[@TokenizedAI]

That's why it's important to binge on a Series 😂

[@mornepatterson6260]

Super high quality and in-depth video. Very refreshing and whilst it takes a bit of time to work through and focus, the content is of such a high standard that it's totally worth it. I've just subscribed to your channel. Thanks and I look forward to more great content!

[@TokenizedAI]

I know it's not the most easily consumable content, but then again, nothing of value is easy I guess 🙂

[@Brian-ej9ew]

You are truly inspiring. A lot of hard work, trials and errors to find the answers. Cannot wait to see part 5!!

[@TokenizedAI]

Part 5 will be a lot easier for me to prepare, but also a bit more fun 😅

[@Rhdefense]

This is excellent information. And thanks for pointing out the difficulties. I made a comment on another video, and you responded. (I’ll go back there and explain myself about the “lately” comment.) I’m now seeing the exasperation I’ve experienced trying to get consistent characters isn’t an unknown issue. It makes it difficult to tell the kind of stories I’m trying to tell. I guess I’ll have to either learn to draw (!) or wait until the AI has advanced. Thank you, again, for your content.

[@Biocodeendeavour]

Thank you for your perseverance and willingness to suffer the frustrations so we dont!! These are incredibly helpful

[@StyleViewStudio]

This is what I was looking for! Wow! Chris - you have no idea how much you give to your audience! (the ones who are trying to learn).Thank you! 🙏PS. Currently anything created with MJ or with prompts," cannot be copyrighted " with USPTO - UNLESS ... it gets remastered with Photoshop or Illustratror . So, all the Artists out there - - brush up your Photoshop skills!

[@TokenizedAI]

I guess the question is how much editing actually needs to happen for it to be copyrighted.

[@bobtzeng]

What a great job and the sharing you have done for everyone in Midjourney world.

[@TokenizedAI]

Thank you 🙂

[@flotantte]

This is really helpful... Let me add another thing. I'm trying to create a little story with little aliens and keeping consistency is really impossible! Midjourney does not understand when we talk about aliens and humans, especially if it is an "alien child", or an adult alien, if it's female or male. The style is Pixar-type which adds another level of difficulty... but without a doubt this has been a huge help!

[@duesenberger]

Very valuable! Thank You! I just discovered that you provide documentation for each video you create, which makes Your contributions all the more impressive!

[@TokenizedAI]

I do my best to create value for people

[@suzevidz]

I've actually had pretty good luck getting consistent multiple characters to appear in my output. I had several drawings that I had done and I used those as reference and I also used the trick that I think you had in an earlier video where you name the character and continue to use those names. Of course, I still get some that are nowhere near what I want, but most of the time it doesn't take long to get what I'm looking for. I think it helps to have the reference images that I drew, along with the character names.

[@TokenizedAI]

There are a wide variety of things that can help with this in certain situations. Tough part it doing it consistently.

[@suzevidz]

@@TokenizedAI Yes, but I noticed a big difference in how much easier it became after I started naming the characters the way you had recommended. Like you did the character Carla. Great idea!

[@TokenizedAI]

Yes, the MJ team has confirmed that this works or at least helps.

[@alicelin1875]

Thank you!!

[@IanHollis]

These are fantastic. Thank you so much for these. I learnt a whole new bunch of stuff about prompt engineering from these, and all for free! You're awesome! I hope in the future I can click on a character and tell MJ to make than character an actor which I can name and use in other prompts. You videos have certainly helped *A LOT*

[@TokenizedAI]

Thanks so much for the feedback 🙂

[@kens.5095]

Incredible resource! Very generous of you, Christian!

[@TokenizedAI]

Just sharing what I'm learning :)

[@SiraEkabutPlus]

Your channel is really get in depth with MidJourney prompt. I really love it!

[@TokenizedAI]

Thank you 🙂 That's the objective

[@Stockholm_Syndrome]

Probably the KZbin channel with the best sound! What kind of microphone are you using?

[@TokenizedAI]

You'd be surprised. I only use a simple Rode NT-USB Mini, but I process all of my audio with Adobe's AI audio enhancer now. It's pretty insane.

[@Stockholm_Syndrome]

@@TokenizedAI -Wow! Well, the sound quality is very good! Keep up the good work!

[@pakmatthew]

thank you! your emails are helpful and now i'm just starting to watch your videos. cheers to you and your successes.

[@TokenizedAI]

Thank you 🙏🏻

[@huibert_devet]

faszinierend, was Du da machst. Wenn ich das noch lerne, ... wie ich einen Charakter als dauerhafte und erkennbare Figur in verschiedenen Umgebungen und Situationen "leben" lassen kann, mache ich Comicstrips für meine bisher nur sehr wenig beachtete Zielgruppe 😉. Ich probiere mich schon an Multiprompts, aber noch sind die Ergebnisse dürftig und nicht planbar. Allerdings beginne ich zu verstehen, wie diese KI "denkt" 😎. Sie ist ja vor allem dann aufgeschmissen, wenn man vom Standard abweicht. Dank noch mal für Deinen wertvollen Input 🙂 hey Guys and Girls, sorry for my English 😎. Next time in English 🙂

[@TokenizedAI]

Ja, im Grunde ist es halt so. Jeder Honk kann man mit 2-3 sinnlosen Worten ein erstaunlich schönes Bild generieren. Aber sobald man eine sehr konkrete Vorstellung davon hat, was man erreichen will, wird's halt echt schwierig.

[@huibert_devet]

@@TokenizedAI erstaunlich schön ist ja schon mal was ... :) aber gewollt schön, da will ich hin 😎

[@christophschmidt1786]

7:30 "This is a collective effort." No, it actually is a technological problem. This needs to be resolved in the software by assigning the reference images to specific parts of the prompt, e.g. by assigning the names of the characters to specific reference images. I am sure it is possible, and will be addressed in future versions of these models.

[@TokenizedAI]

I think you misunderstood what I meant by that statement.

[@christophschmidt1786]

@@TokenizedAI I just wanted to point out that at this point in time, users can only find work-arounds, but the real solution has to be an algorithmic one, or a feature, which connects the reference images with parts of the text prompt. Currently, the computer rather mixes two images than recognizing them to be two separate characters and understanding who is who.

[@SFAutor]

Great video. Your hints have resolved some of my problems I have had. Thank you!

[@TokenizedAI]

You are so welcome!

[@davidmichaelcomfort]

@TokenizedAI adding “film still” to the prompt helps with three or more characters. At least in photorealistic images

[@TokenizedAI]

Interesting, I'll have to try that out.

[@WilliamBraddock-og7oi]

I am loving your tutorial about doing consistent characters.

[@TokenizedAI]

Excellent! Hope you're making progress.

[@michaelsbeverly]

You're doing god's work here..hahaha....thank you. I've had many nights this week working until 2 or 3 am as if I was in Las Vegas pulling a slot machine handle wishing for a jackpot....one thing I don't yet understand is why, when you have a great scene you love, but want to adjust slightly, the adjustment takes a perfectly fine face and turns it into an abstract monster face....

[@TokenizedAI]

I think I'll leave the answer to that mystery to the MJ team. I have absolutely no idea and agree that it's difficult to comprehend why this happens. PS: I love the slot machine analogy! 🤣 It's so true.

[@jaytamati4120]

Thank you so much for this. I have just started and watching this has helped me so much.

[@TokenizedAI]

Glad it was helpful! 🙂

[@yveskeroack1612]

Great stuff again Christian! I also noticed how easily concepts bleed in MJ, which is a great feature if blending images and concepts is what you want to do, but becomes problematic in situations like what you just described. Maybe MJ needs something like a -bleed argument that would help control this in a similar way -stylize works, with a range of values and a set initial value ;-)

[@TokenizedAI]

To be honest, bringing back image weight (--iw) for v4 would already be a huge improvement. The lack of control over how much influence the image prompt has is a bit annoying and a major limitation. It also makes all these tutorials a bit obsolete once we get that control.

[@thebluriam]

"multiple characters require strict archetypical contrast". That's a huge key idea right there!

[@TokenizedAI]

I was actually thinking the same thing when I wrote it down 😂 But kidding aside, it's one of the key takeaways. Not sure everyone understands it though.

[@thebluriam]

@@TokenizedAI yeah, it's kind of a strange abstract one in a sense. Very difficult to communicate because it requires someone to kind of "get" how images in the training data "work" and how they relate to each other. I guess we could say to people, "The AI is like a child whose imagination sometimes dreams up weird things because it's too young."

[@TokenizedAI]

@@thebluriam You forgot that people actually need to understand what an "archetype" is, too 😆 And I suspect far less do that we might think. I only realized this in hindsight.

[@thebluriam]

@@TokenizedAI oh yeah that's true. Dang, lots of work ahead of us, thankfully we have AI to help now lol

[@RasmusSchultz]

ControlNet ought to make this kind of art direction a lot easier - I wonder if that's coming to MidJourney? They would probably start to need a proper GUI at that point. 🤔

[@TokenizedAI]

We'll see what v5 has in store.

[@carlosbenitez6438]

hands down the BEST series and channel by a long shot. SO INCREDIBLY HELPFUL. THANK YOU!

[@TokenizedAI]

My pleasure

[@JimArthurVanWyck]

Wow…. Best MJ tuts ever… brilliant

[@TokenizedAI]

🙈 Thank you so much 🤗

[@dreamdesignproductions9246]

I love these videos. Very detailed and gets to the point. Thank you for putting in the effort to be able to teach us these amazing tips and tricks.

[@TokenizedAI]

You're very welcome!

[@benzyn]

Brilliant video as susual, Christian ❤....thank you for sharing these excellents tips for free

[@TokenizedAI]

My pleasure 🙂

[@neuralpunks]

nice! Can't wait to plug this method in!

[@TokenizedAI]

Good luck! :)

[@EdmundShern]

Great format and very educational! I hope you do more prompt engineering tutorials/ explorations like this!

[@TokenizedAI]

The channel should have a whole bunch of them

[@terrestrialx7335]

Thank you for all your help. Just learning and love it. Wondering how? Where? you get your image URL's thanks

[@TokenizedAI]

Right-click > copy image link? 🤷🏻‍♂️

[@sceltadelcamper]

So what is your suggestion in order to create multiple different scenarios with 2 different characters and one animals? Is it better to create first the single characters and then try to create a composed scene with all of them or start directly with a complex scene like the one you created with the pool? Thanks

[@TokenizedAI]

Honestly, it's just easier if you use an image editor to mix and match as you need, if you know how to. Most people don't, which is why it such a requested solution.

[@sceltadelcamper]

@@TokenizedAI I have the same idea, but having midjourney at our disposal, it is natural to try to create the scene entirely with this tool. But it is not so easy to maintain consistency. I'd like to support you and your channel in some way, please let me know if there is a sort of subscrition. I also have a question: you spend a lot of Time to create learning content for the channel, but are you working at the same time on projects in which you use your AI knowledge? Thanks

[@stellar-pix5476]

can you provide a list of the most current videos on character development in order? I just found your videos (awesome by the way) and you mention #2 is a series, #3 etc but I don't see them on the titles. I would like to watch them in the order that each subsequent video is building on the prior - Thanks Great stuff Christian!!

[@TokenizedAI]

Check the character design playlist on my channel. It's also linked in the description.

[@TheTimedCarbGuy]

If you would mark the things you added in the current prompt (in regards to what's been there before) it would be easier to follow along when the prompts get longer and it would be a little more fun to look at the text prompts :) The way the bear and kid pictures turned out was realy great. Is it different in MJ 5? Is it easier to put multiple characters into a scenery?

[@MarySmith-sg1ns]

Can I use BlueWillow instead in this process?- Please share a tutorial using this tool as well. This was amazing insight! Thank you for taking your time to explain all these approaches

[@TokenizedAI]

Unfortunately, I'm not going to do anything that supports BlueWillow after they've been spamming all the KZbin channel comments with orchestrated campaigns that flooded them with fake comments. They dropped the ball the moment they chose to use such shady tactics.

[@JohnVanderbeck]

Honestly it is things like this that drove me to stable diffusion webui. I still do a lot of ideation in MJ but I always finish in SD where I have about 5 bajillion percent more control

[@TokenizedAI]

And yet, everyone agrees that the output from MJ is just so much more polished. Also, Stable Diffusion just isn't very accessible to the majority of people out there. I know that those who use it don't understand that because it seems easy to them, but trust me, Stable Diffusion has massive issues in terms of user-friendliness and barriers to entry. As with nearly all open source software.

[@Decora_con_Cris]

Could you explain please why you use 2 segments in the promt with the little boy and the bear? I got lost there. Thanks for everything you’re doing in this channel! Real value!

[@TokenizedAI]

When you only use a single segment, there's no real guarantee that your prompt will be interpreted that way you believe it should be. By describing the exact same thing with an entirely different prompt (as a 2nd segment), you reinforce the concept. Midjourney tries to interpret both separately and then blends them together into a single image. If one of the segments is off base, the second segment increases the likelihood of actually getting what you want. You can add a 3rd or 4th segments as well to reinforce it even further. This method is especially useful when you have 2 or more subjects. If you only have a single segment, there's always an inscreased risk that MJ will not show them separately and turn them into a blended character instead.

[@Decora_con_Cris]

@@TokenizedAI thanks a lot! It’s clear now!!!👏🏻👏🏻👏🏻

[@budi_hartono]

You could try adding the --seed parameter to keep results consistent on Midjourney

[@TokenizedAI]

Yes you could. But setting a --seed only makes sense within a very narrow scope of your prompt. It also assumes that you want you have already found a specific composition you like. If you haven't, then it's counterproductive. And with regards to the first 2 failed attempts, the --seed does nothing to help in that respect. The problem lies elsewhere.

[@arialopez1983]

SO COOL 🤩 Please share a tutorial showing this process but using BlueWillow? You would be a lifesaver

[@TokenizedAI]

I haven't experimented with BlueWillow very much yet.

[@AIGeneratedVlogs]

Thank you for this tutorial, Going to practice now.

[@TokenizedAI]

Have fun!🙂

[@bettymac8248]

total gem prompt: "Character concept on white background, multiple poses and expressions" GAME CHANGER!!

[@cn6519]

In other words, just hire an artist to build the illustration, or cut out all of your images generated on mattes, and then comp image program like photoshop.

[@TokenizedAI]

That's an option, yes

[@jayc6159]

Thank you that filled in a couple of gaps for me.

[@TokenizedAI]

Awesome 😃

[@Great.AnotherChristian]

You can speed up the process by making your characters a preferred setting (example:: --peggy --carla)

[@TokenizedAI]

Not really though? Because custom options only accept parameters, not text prompts.

[@Great.AnotherChristian]

@@TokenizedAI if you check out some of my Midjourney generations then you may see something I didn't. I get pretty consistent pictures of "Sorrkin Ironmaw" in particular.

[@Great.AnotherChristian]

@@TokenizedAI if you have the time, that is.

[@TokenizedAI]

I was just going off the official Midjourney docs, which suggest that you can't add text prompts to custom options. I actually thought it gives an error if you try to. Are you saying it works on your end?

[@harrygoldhagen2732]

excellent as always! It seems like the best way to combine characters and backgrounds is to make them separately and then combine in PS.

[@TokenizedAI]

Yeah, for the time being it is

[@healthylife8820]

Great work! Thank you for taking your time to explain all these approaches.

[@TokenizedAI]

My pleasure!

[@luiseterovick7096]

This video is pure gold

[@TokenizedAI]

Thanks 🤗

[@Ben42-Ben42]

Thanks. That series is really helpful. I tried it by myself but did get it right. So: tanks🎉

[@TokenizedAI]

My pleasure! 😊

[@Hassaan911]

This is the best solution ever I found... big thumbs up 👍

[@TokenizedAI]

Glad to hear that 🙂

[@lych1669]

One of the best video series about MJ! I'm having a question however. Why is the seed used for Carla at 1:16? And would it also be possible to just use the reference of Carla's basic reference picture instead? I don't quite get what the seed is good for here.

[@TokenizedAI]

Because I wanted to recover the exact same image the I used in a previous video. I need to use a seed for that. I was just showing how others could retrieve it as well.

[@jasonrist6582]

owlslothbear cant hurt you. owlslothbear doesnt exits. Midjourney: BET

[@TokenizedAI]

🤣🤣🤣

[@Animadroids]

What does --seed stand for in the prompt ? Is it an Index of a previous photo result ? Does it persist many sessions? Thanks

[@benmowlah8943]

Hi. Apologies if you have already answered this - so how do we insert two characters into a scene, rather than generating a scene with new characters? ie was there a solution for Peggy and Carla in attempt #1? Thanks

[@TokenizedAI]

I didn't answer it directly but I though it might become clear from the video. Basically, you have to completely change the way you prompt in order for it to work. Hence the demonstration in this video. Using the old Carla the way we created her before isn't suitable for creating multi-character scenes. This is simply a limitation in how all these AI generators work. Your best best is to create characters individually, cut them out and then place them into a separate scene with an image editor. The tech simply isn't ready yet for these complex use cases.

[@benmowlah8943]

@@TokenizedAI Thanks!

[@freshlesh3019754]

Just a question, why not use the character design seed to inform the generated background style and then overlay them? 6:46

[@TokenizedAI]

The seed won't make any difference unless the text prompt already gives you the right concept. If it does, then you can set a seed, sure. But in the end, your concept needs to work. Setting a seed won't fix it by itself.

[@GAllanRoberts]

Thank you, for your research and presentation. I learned a lot and I can wait to apply it to my prompting.

[@TokenizedAI]

Wonderful!💪

[@malhayn]

adding --seed --stylize @ 16:17? Just wonder if using a --seed referenced to a previous render of boy-girl-rabbit-pool ALONG WITH the two image references could help maintain character consistencies: hair, clothing (style & color), as well as, introducing a lower level of --stylize 50? to reduce the Midjourney induced variability within the new renderings? Any ideas? Wish I had the time to play with Midjourney. You do good work! Thank you. ( --stylize per Midjourney: Low stylization values produce images that closely match the prompt but are less artistic. High stylization values create images that are very artistic but less connected to the prompt.)

[@TokenizedAI]

Honestly, I'm not a fan of --stylize. And using a --seed really only makes sense in very specific situations. I generally don't like to tweak the default settings too much because it makes it very difficult to determine what exactly produces different results. And the vast majority of people use the default settings, so I want to stick as close as possible to what your average user is doing.

[@malhayn]

@@TokenizedAI Makes sense. When I get time I'll run some test and share any positive results. Thank you for making everything so understandable.

[@TheCubicplanet]

Have you tried to use different animals but also describe their position in the image, let's say from left to right? Maybe it helps the algorithm and prevents it from mixing different animal traits? Just a thought. I'm going to try this myself right now.

[@TokenizedAI]

Yes, the main problem seems to arise from animals that are completely unrelated and thus very little photos exist that show both. For example, cats and dogs work quite well, but honey badgers and rabbits do not.

[@luisoares1991]

I must say your results were impressive! Have you tried using the seeds of the image? I believe this along with your script and the link to the reference image could help a lot!

[@TokenizedAI]

I have, but in this case it would have conflicted with the references.

[@thembasimelane]

Thank you sir, I been dying to learn these!

[@TokenizedAI]

That's awesome to hear!

[@yeastydynasty]

Hopefully that gets fixed soon. Doing comic books with Midjourney is difficult as it is and it sucks that you can have two different consistent characters in one frame.

[@TokenizedAI]

Remember, the tech is still early. You could barely do any of this only a few months ago. Let's be patient and see where the journey takes us.

[@adsertheblade]

Awesome, but does it also work with Carla and Peggy? I'm managing to keep my dog consistent, but not the human...

[@whyfollowmenow]

IDK if it's been mentioned, how about shuffling the order of the prompts? assuming the image starts from left to right and top to bottom. Ie should the characters attributes be mentioned first or the background, or the style.

[@TokenizedAI]

Doesn't really make a difference. Sometimes you'll get something that looks like it made a difference, but it's just random luck. Our confirmation bias fools us. If you keep re-rolling you'll notice that it doesn't really do much.

[@SuccessTakesPerseverance649]

great video very enjoyable and interesting is it possible to get the prompts you have used

[@TokenizedAI]

Thanks! The prompts are all in the video.

[@ozmenlaps3038]

U are pure gold, sir! Thank you for your time and hard work! You are making my gurney to my own world a lot easier) 😍

[@TokenizedAI]

Happy to hear that!🙂

[@GUARDONISSIMO1]

Hey Christian, thanks a lot!! Beautiful series and videos in general, from editing toy the shots, to your background... May I ask: are you based in Belgium or ermany?

[@TokenizedAI]

Thank you so much 😊 I'm based in Hamburg, Germany. Beautiful city by the way! ⚓

[@GUARDONISSIMO1]

@@TokenizedAI Ich weiss.

[@andyrevo8081]

This is the University of Prompting, by Prof. C.P. Heidorn (middle initial P. stands for prompting)

[@TokenizedAI]

Hahaha 🤣

[@HitsandHeadlines]

hopefully mid journey 5 will solve these issues

[@TokenizedAI]

I think a lot of things will be solved in v5.

[@brian7android985]

Interesting that you can see influences where the data sets are sparse. E.g. several cute bears had obvious traits taken from Poo Bear.

[@TokenizedAI]

Yeah, I think there is a ton of obvious bias in the dataset. However, it gives us an idea of what you could do if you trained the model for a specific use case.

[@00Mass00]

Very cool! I'm a newbie, so how do you get those image reference links? Save as, and then reuploading in Discord?

[@TokenizedAI]

If they already exist on the web, then just copy the image URL/link. Otherwise, yeah, you upload them in Discord them and then do the same thing. If you need one of the variations, react with the envelope icon and then grab the individual URLs from the DM you receive.

[@pineconey]

Great guidance--thanks for this!

[@TokenizedAI]

You're very welcome!