Leonardo AI Are Consistent Characters Possible?
Vložit
- čas přidán 8. 06. 2024
- Leonardo AI Are Consistent Characters Possible? Today I want to show you some simple techniques to get character consistency. Although it's not 100% accurate we can get fairly close with some techniques. By nature AI generated images are random, so controlling the outcome can be a little tricky. This video is for people who can't run stable diffusion locally and have to rely on online AI image generators like Leonardo AI. leonardo.ai/
Prompts: 2d character turnaround multiple action poses, Happy joel quero bruce lin kpop star, dark brown hair blonde highlights, fashion pose, black t-shirt, blue jeans, average build, clear detailed eyes, fine details, sharp focus, reflective eyes, digital illustration, white background
2d character turnaround, fashion pose, anime spanish-filipino lisa tison katty dobrev, female wearing a plain blue t-shirt, beautiful face, fit build, high details, digital illustration, white background, Artgerm, WLOP, tom bagshaw, magali villenueve,
Video on ControlNet • Leonardo AI Levels Up ...
⏲Time Stamps
0:00 Is character consistency possible?
0:30 How to get consistent faces
3:36 Character turnaround
4:42 Use upscale options to recover details
5:16 Decide on a trained model to use
7:02 Inconsistency can work in your favor
8:00 Use ControlNet for posing
8:45 Use seeds
*Disclaimer Affiliate Links Below*
📸 Gear I use
Sony A7C CAN amzn.to/3spWX8C
Sony 35mm 1.8 CAN amzn.to/36Apekr
OBS (FREE) obsproject.com/
Editor Davinci Resolve www.blackmagicdesign.com/ca/p...
Audio: Rode Podmic amzn.to/35sSnxv
🎵 Epidemic Sound
🔦 Find us on:
Discord: / discord
Instagram: / monzonmedia
Facebook: / monzonmedia
I know I say consistent a lot! hahaha! This video are for people who can't use stable diffusion locally. Obviously you would have more control and options in Automatic1111 however, these workarounds can get you very close. We will go in-depth on this in future videos to come.
Super cool Ermin! Thanks for your detailed information on this great tutorial.
Hey Chef! Great to hear from you, hope all is well! Would love to catch up soon. Thanks for stopping by! 🙌
Thank you for these tips, I have been having a hard time getting consistent characters, I will try this out today.
Keep me posted on how it goes, again I want to stress it's not a fool proof method and more of a workaround. It may still require some post production.
Hi! This was the first video I watched on your channel; it was so good that I subscribed. I've been watching several images with characters side by side and had no idea how people were doing it. But, honestly, I did not search about it because I was trying to learn about more trivial stuff. Last August I tried Midjourney and then gave it up. So much has changed since then! I understood what you explained and I will try it soon. Thank you.
Thanks so much for the support and kind words! I’ll be covering more on this topic however unfortunately not with Leonardo AI as I’m now working with Playground AI. That being said, most of what I teach applies to any stable diffusion platform 👍🏼
@@MonzonMedia ,yes, Leonardo, BlueWillow, Playground are several apps that I found. I haven't tried these two yet. I imagine that each one has its specificity but the base will be the same. By the way, if you can tell me, do you like Playground better? Right now I'm exploring Tiling in Leonardo and I haven't been getting satisfactory results. I can always improve the image in the PSD but I wanted something perfect and I'm not getting it. I didn't see very good patterns either, maybe it's not Leonardo's forte, as other types of images seem to come out very well
.
Just started with AI art, and this video is very helpful! Didn't know I could do the same character but in different poses. Thank you mate
Hey Tyler, appreciate the feedback and glad you found it helpful! Keep in mind since this is based on stable diffusion, it applies to any other platform. 👍 You can also utilize seeds in a similar way.
I subscribed because I like the facts that you are very articulate and do not speak down to us. Please report, immediately once somebody accomplishes true character consistency - can't wait!
I have something planned however it won't be for Leonardo, just stable diffusion in general which can still be useful. Thank you for the kind words and support! Means a lot to me. ☺
Another amazing tutorials! Leonardo and Playground is currently what i am learning right now and this is helpfull!! Thanks again for these videos!! May success be with you 💪🖒
You are too kind my friend, thank you! More to come soon!
@@MonzonMedia will look forward to it
Thank you so much. Your video helps me a lot.
Glad to hear that!
Great tips and will definitely try those out!
Hey dude! Always a pleasure to hear from ya! Let me know how you make out on the Discord 👍
Hi @MonzonMedia, thank you very much for the link and for the explanations, as I see you in my other channel I just made my inscription here too today, sorry for the delay. I will try to learn with you and I will try again, but now I will do not with Mage but with Playground AI. And I do not know if I already said, anyway if I did I'm doing again, congratulations to you to get a position in Playground, is more than fair as you have been doing a terrific job with the videos that you already record for us, showing what is possible to do with Playground.
You're very welcome Jose! Thank you so much for the kind words, I'm having a lot of fun with PG but working even harder now! 😁 These techniques can work in PG but it's bit more trickier to get the turnaround right. Play with different sizes as it affects the outcome as well. I will do a video on this for PG eventually. Thanks again for your support! Means a lot to me. 🙌👍
Thanks, I'm looking forward to trying this out
You're welcome! Let me know how it goes!
Hi @MonzonMedia, Amazing what you can do with Leonardo AI, thank you very much for share your work with, I'm learning a lot with you.
Hi Canal, I'm glad to hear that! I have more on this to come so stay tuned! I appreciate you my friend! 👍😀
This video 's very helpful ,Thank you.
You’re very welcome! Glad it was helpful. 👍🏼
This was very helpful for me, it was still very inconsistent at times at least that’s been my experience and maybe its the style I’m working with but it doesn’t want to give me a character sheet at times. Then when I do get a character sheet it can be somewhat consistent. Not sure what I’m exactly doing wrong BUT I was able to get enough for what I’m working on. Which I’m thankful for, best part is gaining the experience and learning and getting better and better
Hi Heritage Joy, yeah it's not a fool proof method and because stable diffusion is so random by nature, it will be inconsistent with every generation. The tech is advancing fast though and I think we will be closer to being able to do consistent characters a bit easier soon. There are models you can run locally that are trained for character sheets so that's an option if you can run it locally. But even then it's not perfect either. What I'm experimenting with is doing a few that are close enough, then using canvas on Playground AI to make adjustments and changes because it has more "editing" control. For this type of character creation it will take some post production.
First time doing a.i and i stumbled upon your video... Now i am so pumped!
Glad to hear it! Are you only using online platforms or are you running it locally?
@@MonzonMedia just the online platforms as of the moment....
A great video ❤
Thank you so much for taking the time to watch and comment! 👍☺
Leonardo AI itself should be producing tutorials how to do these things. Until now we are using trial and error to get to what we want.
Definitely agree and from what I hear they will have a dedicated channel very soon. With that being said, the concepts apply to any stable diffusion platform so if you learn about stable diffusion in general and how it works, you will be able to use any local or online platform. 👍
I’m really struggling with a kids storybook I’m trying to create. The character is not a human, it’s a fish. But trying to get consistent back side & front views is a nightmare. Are there any tips you could give me? I use Leo & playground mostly. Only on iPad tho. Thanks for your awesome vids, and I’m glad you are now doing playground videos too 🇦🇺❤️
Hi MissChelle, it can be tricky at times and will most likely require multiple generations. You can try something like this prompt: character sheet, multiple poses, cute fish, cartoon, pixar. Or use character turnaround insstrad of character sheet, it also might help to specify front view, back view, side view either together or individually. Using prompt magic Leo helps with this.
Personally I've found the easiest way for specific positions is to use ControlNet with a reference image, either free hand drawing or a stock image. Hope that helps! 👍
Thanks for the video
So long story short, nothing beats training models :) Find on internet 15-20 images of a character (or just one 3D model and create different images of it) and train a model with it. You'll spend more time but the result will be much better
Definitely nothing beats proper model training however it still doesn’t address getting consistent results with clothes. Training Loras is the closest we can get to this point. Then there is using ControlNet reference that works really well without having to train. Furthermore, I think the way things are evolving, there really shouldn't be any need to train models to get consistent characters. This is supposed to be easier right? We'll get there eventually.
That was VERY helpful, thank you! Do you know of a prompt that would avoid background while creating, something like a green screen?
You're welcome, just put in your prompt what you want, plain white background, solid green background etc. Might have to try a few and see what works and it won't always give you exactly what you want. A few re-rolls may be needed.
I definitely planned to create character designs based on ebooks I wrote.
That's awesome! I hope this helps a bit. I've also wanted to do a comic and short stories for some time. I think I will make an attempt sooner than later. 😁
Great video! How do I feature my consistent character in different scenes, such as at the park or at school?
Hi again, how do we feature two or more consistent characters in different scenes? Thanks!
cool
🙌Appreciate it!
ciao bellisssimi i tuoi video, grazie. vorrei chiederti una cosa, possiamo inserire un disegno creato da noi sulla maglietta delle modelle Ai?.
Thanks for this! Very helpful video. Question though, what prompt do you use to get a full body shot always? I use “full body shot” but at times I get a close up or just a half body image.
Best to give more context, I like to mention articles of clothing like jeans, converse shoes and even stating what they are standing on for example walking on the sidewalk or standing on a mountain. Works every time.
@@MonzonMedia Alrighty I shall try that. Thanks so much.
Very helpful. What if I have want to use an existing picture? For example, I have a headshot photo, and I want to generate a full-body image, and then use that as a consistent character (able to change clothing, poses, etc.)?
That would require training a model which you can do in Leonardo however the free version you are only allotted one model. czcams.com/video/oNajDvx3_GY/video.html
I guess that blonde model at about 5m.30s mark was cold hahaha! The amount of times I was typing an answer to a supposed problem, you solved it before I could finish! You should rename to the A.I Guru haha.
😁OH geez, I didn't even notice that hahaha! Dude, this was originally a 20min video so I was worried that I cut out too much. I will dive deeper on this on videos to come, there is WAY too much to go over in one video. I hope this was a decent enough summary. Using the Prompt Magic V2 helped a lot with the turnarounds. Appreciate you stopping by!
@@MonzonMedia Always a pleasure to visit your house! I think you got it down perfectly tbh. Like I say you answered any questions I was about to try to clear up. Also you now have a pt2, training those character sheets into a trained model. Keep this great content coming!
Glad to hear that bud! Yes Part 2, part 3, part 4 to come....I have notes upon notes to go over!!! 😁
Hi, those were some pretty informative tips, but I am having trouble with some other settings of the tool, such as:
1. Changing the number of step counts
2. Changing the sampler to Euler Discrete and others without changing the finetune model and finally,
3. Changing the base model.
The options are just not there, any ideas why? Maybe I need to get the paid subscription for that?
BTW, thanks for the cool video!
Hey there, yeah there are some limitations to the free membership so points 1 and 2 are because it's a free account. As for selecting a model you just have to go on the drop down and select "custom model" you can see how here. czcams.com/video/ii_fEY_e5nQ/video.html Hope that helps!
@@MonzonMedia Ah, I see. So that's what's going on. Thanks a lot for getting back to me and shedding some light on that.
Where is the perfect place to paste the seed? At the beginning of the prompt, or at the end? Thanks!
Under advanced settings on your left menu there is a dedicated box for seeds to be entered. Toggle on "use fixed seed" and enter it in there. 👍
Is there a way when making a coffee maker via AI to generate different views of it with the same details ?
Hey Claas, I would think that if you use the seed number you can get some consistency but because by nature stable diffusion is random, it can be pretty tricky. It will most likely need some sort of post production afterwards.
@@MonzonMedia Thank you - that is what I assumed.
But how can I use my own character face. I have only one image, and I would like to use that same image to generate different poses, and angles. I don’t want the AI to come up with its own face.
To do that you will need train a model and you need minimum 15 images but 25 is ideal. czcams.com/video/oNajDvx3_GY/video.html
Does this work with animal characters as well - I tried but didn't get as consistent results
It can work but as I said in the video, best to keep it as simple as possible but the fine details are hard to control and make them consistent. By nature AI image generation is random, although there is new tech coming out that will eventually help consistency. Most of the work arounds will have to be done in post but we're not far off from being about to do it easily.
how do you get multiple character turn around, from one single pose?
I'm not sure I understand your question, can you clarify? In the video I mention using "character turnaround, Character sheet" along with multiple poses, or even specifically front, side, back views.
@@MonzonMedia i want to create a character turnaround for a pre-existing character I've already designed.
Is it possible to multiple poses in ControlNet with the turnarounds?
Great question! Technically yes but it has it's downsides as well. You spoiled one of my next videos! hahaha! I've found that it can take some work to do BUT well worth the effort. Basically what I tried was using my own photos and creating a template in poses that I wanted however keeping clothing consistent is a challenge. I wanted to do a few more tests before I do the video on it. But you are on the right track. 🙌
@@MonzonMedia I thought that might be the case, both that it might have some downsides and that you might be doing a future video on it. So now I have something specific to look forward to in one of your future videos.😁
How Can I find ur works in leonardo community feed
Hey there! Unfortunately all my work has been privatized, I took a job with playground ai a few months ago and due to a conflict of interest I can no longer cover Leonardo AI, or any other online competitor. However I'm still covering stable diffusion content just on local platforms but you can still learn the same concepts as they do apply to the online platforms. Hope you decide to still stick around. 👍 I will be updating my main website soon, which is severely outdated and will be posting some of my AI work there.
@@MonzonMedia
Thanks sir
I can't believe just what the hell happened: The AI started spitting out full graphic novel pages with my prompt, although most are bad because of the resolution. Even then, some are actually viable and only need text bubbles. After that, I started pumping out good characters with this prompt:
2d character turnaround multiple action poses, ((A masterpiece)) (cyberpunk dystopia) full body portrait of a Cyberpunk 2077 rugged Vietnamese-Russian woman (wearing plain navy blue security officer uniform), (hyper-realistic) color pencil portrait, Banksy art, Kim Jung Gi, freedom, soul, digital illustration, comic style pinup, steampunk noir, approaching perfection, dynamic, highly detailed, watercolor painting, Art Station, concept art, smooth, sharp focus, illustration, art by Carne Griffiths and Wadim Kashin, Bladerunner theme, futuristic and dystopian, cinematic, dramatic lightning, full colors, 3d, render, dark, high contrast
Negative: (((2 heads))), horns, duplicate, blurry, abstract, disfigured, deformed, cartoon, animated, toy, figure, framed, 3d, cartoon, 3d, disfigured, bad art, deformed, poorly drawn, extra limbs, close up, b&w, weird colors, blurry, watermark, blur haze, 2 heads, long neck, watermark, elongated body, cropped image, out of frame, draft, deformed hands, twisted fingers, double image, malformed hands, multiple heads, extra limb, ugly, poorly drawn hands, missing limb, cut-off, over satured, grain, lowères, bad anatomy, poorly drawn face, mutation, mutated, floating limbs, disconnected limbs, out of focus, long body, disgusting, extra fingers, groos proportions, missing arms, mutated hands, cloned face, missing legs
Dreamshaper v5, Leonardo style, 640x832, Guidance 7, Seed 228459776
im having this problem anyway if it can be done now
Still the same really, the tech isn't there just yet. If you are able to run Stable Diffusion locally, there are some work arounds but there will still be some minor post production required.
@@MonzonMedia i’ve read on Leonardo forums theres a thing called Training & Datasets could that add consistency i also read if you put the image seed into the prompt you get more consistency i’ve run out of tokens for today i’ll try tomorrow . its annoying i want to start making visual novel. I think in a few years you be able to transport the image into a movement prompt but the technology is still young
Your tutorial is difficult to understand. It feels rushed. I couldn't learn anything. I have watched it many times but no result
What are you trying to learn? If you are just starting out with text to image this isn’t the video to start with. Some context would help so I can understand what you are struggling with.
@@MonzonMedia I am struggling with image consistency for a cartoon character.
How familiar are you with prompting? Do you know how to structure your prompt and get the desired image to start with?