Playground AI Consistent Character Hack No Training Required

Playground AI Consistent Character Hack No Training Required. So I scoured the net finding this work around for you all and I think it works fairly well with a little patience. This method can be used with any stable diffusion platform but I will demonstrate it using playground ai.
Be sure to use these negative prompts! docs.google.com/document/d/15...
For female characters add to negative prompts: Male, man, men, boy, guy
For male characters add to negative prompts: Female, woman, women, girl, lady
Check out Playground AI here, for a free ai art generator (also a paid plan) it's of great value!
Site playgroundai.com/
YT Channel / @ucublzj30gle1f5oswnoys9g
⏲Time Stamps
0:00 Overview and how this works
0:48 Tweaking your prompt
2:06 Settings
2:43 Generated images and what to expect
4:44 Using filters for variations of your character
6:44 More examples, and male characters
*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

[@MonzonMedia]

So I should mention that the main reason to do it this way "without the filter" helps generating the consistent characters. Technically you can use 2 made up female names + a filter like perfume and still get consistent characters. Either way you do it, at the end of the day you get a consistent character. I haven't seen anyone else approach it this way.

[@dennyrullytahier6239]

Amazing! Thanks for the tutorials, i'm still new and still learning and this helps a lot!! Thanks a million for this!

[@MonzonMedia]

You're welcome and glad it was helpful! 👍 Check out this follow up video, although it's on a different platform, the same concept can be applied. kzbin.info/www/bejne/f4GtgoJrqNOAmpY

[@dennyrullytahier6239]

@@MonzonMedia sure thing, checking it right now 🖒

[@MonzonMedia]

**How many times did I say charachter? hahaha!*** Stick around for the version for Automatic1111 or Invoke AI. I'll cover doing the same process as well as using free models from Civitai and a trained model from Photobooth.

[@luisdelarosa-sabina1656]

I cannot wait, friend. I will be looking forward to watching that upcoming video. Please, and thank you in advance.

[@zacharysherry2910]

It's spelled cheraptor

[@JimArthurVanWyck]

Great video. Thanks so much. This is exactly what I was looking for

[@MonzonMedia]

Glad to hear Jim! Something to consider that I think works ok too but I didn't really test it out too much. If you use fictitious female names + a filter you can get pretty consistent results as well. Try both! 😁👍

[@Grimmona]

Thank you so much i'm loving this! One little question, wouldn't it working better, with just one actor or actress?

[@MonzonMedia]

Great question! Basically you want different traits and characteristics to mix otherwise they can look too "manly" in the case of woman characters. Also as I mentioned to the others, technically you can just use made up female names plus a filter like "perfume" since it's based on a model. You can get decent results too, however the method in the video works very well when you are just using your own prompt, no filters. I know it sounds counterintuitive but it works well from what I've seen doing other methods. Hope that helps!

[@milestrombley1466]

Thanks. I been having trouble with Playground AI. 😁

[@MonzonMedia]

Always glad to help!

[@mireauxrascian]

I remember when I was photoshopping portraits in the distant past. I had to manually remove blemishes, smooth the skin through manual techniques, then add back skin-like texture. Now, I use tools for 80 percent of the process. I hope we get to that more automated part sooner or are able to have these ai tools understand us better because no one will pay us salary or piecework compensation to spend an hour rolling the dice hoping to get one image of what we want with these tools. Oh and before you say anything, yes I would pay for that from my own pocket.

[@MonzonMedia]

We're definitely not far off from there. I will say we've come a long way with those kinds of edits. Programs like PS, Luminar have adopted AI in the programs making retouching as simple as sliders but yes I do see some sort of automation happening very soon. AI is moving so fast, it's both great and scary! I appreciate the feedback!

[@fictionaddiction4706]

@M Rascian I still do this, using the clone tool to remove blemishes and adding skin like texture. can I ask what do you mean by using tools to do 80% of the process?

[@mireauxrascian]

@@fictionaddiction4706 Well, it depends. If you are editing high resolution high pixel count photos with face closeups that will be used for print ads or as large website backgrounds, you still probably need to do things the way you do them. However, for everything else and trust me, there is money to be made in other use cases too that people pay for, there are tools like Imagenomic Portraiture or PortraitPro. Hey, maybe they work for those high resolution use cases too.

[@fictionaddiction4706]

@@mireauxrascian Thank you so much for the info, I'll check them out.

[@ShopVajero]

Great video. Thank you so much. I think: the key here is "patience", right? Could you make 1 more video on how to create a character in different poses and expressions on Playground? (I want to use for my story book and video). I mean, if it is necessary to train the AI to create characters faster, how to train?

[@MonzonMedia]

Thank you! Yes patience! I actually have that very topic in line to record soon. Basically you have to put the expressions in the prompt however poses can be a bit tricky. There are key terms that I've been testing, like fighting pose, dynamic action pose, crouching...things like that. The AI is already trained so it's not like it's learning with each generation. We just have to figure out which "words" will give us the best results. 👍

[@ShopVajero]

@@MonzonMedia I'm your fan🤩

[@MonzonMedia]

Wow I have a fan! hahaha!

[@GnanKumar]

I am too waiting for the same

[@ArtBuddy-bf1is]

Sorry it is working again Thank You. You are the Best.

[@dirkbaeten6949]

This is quite an unexpected technique and it's certainly interesting. Well done. For my purpose of creating a visual novel, it would be nice to have a way to keep not just a similar face but also the exact same clothing and hairstyle.

[@MonzonMedia]

Hi Dirk, yeah it's a bit of a strange method. I did fail to mention though that this technique was mainly for the default stable diffusion 1.5 model not using any filters. I did some further testing to find that using a filter will give you those consistent results as well so no need to do the reverse prompt trick. You can enter celebrity names but I find that certain ones still have a very dominant look that it's still noticeable. You can use fictitious names too. I'll be doing a follow up video very soon. As for consistent clothing that is a bit tricky, there are things we can use like seeds to get similar results however the nature of AI is very random. At the end of the day there are more options to create different characters and that's great in my books! Appreciate the feedback my friend!

[@mrbabyhugh]

Just give the character a unique name, then add female superhero, then descriptions. Then always use that unique name.

[@MonzonMedia]

Yes definitely! I actually go over that in this video kzbin.info/www/bejne/f4GtgoJrqNOAmpY 👍Always good to have different options and methods but the purpose of this technique to get particular traits and characteristics.

[@goldencharms1]

What AI app allows you to use a photo then generate consistent characters in a cartoon look?

[@Rherman1950]

Great video, learned something new.

[@MonzonMedia]

Great to hear Richard...it's definitely a strange method but it works! haahha!

[@jasonlee789]

Is there a way to organise my images into separate folders, as it becomes so confusing having to scroll through my images? Thanks

[@MonzonMedia]

Unfortunately not at the moment. You could use Canvas to help organize your images, there is a folder system for Canvas.

[@BezalelRaz]

I am from Israel and I use Playground V2 software to build an AI model that will help me create caricatures of influencers and opinion leaders in the Israeli media. However, the existing basic models of Playground V2, such as Stable Diffusion XL and Automatic 1111, do not recognize Israeli celebrities. I would like to train Playground V2 to recognize famous Israelis and create caricatures in the style of newspapers. Can you suggest how I can achieve this?

[@fccnx01]

Amazing tutorial! But I'm wondering if this "merge" technique could work in Midjourney?

[@MonzonMedia]

Hey Fernando, actually it does without having to do the negative prompt trick. Since MJ doesn't necessarily give you "celebrity" looks it is much easier to get a consistent face with 2 names, even random names. So spoiler alert, this was going to be my next MJ video hahaha!

[@RoboMagician]

about the celecbrity prompts like henry cavill and robert junior, does it have to be western/hollywood mainstream celecbrities. does asian celebrities work at all? or are celebrities from other ethnicities in the data base at all?

[@MonzonMedia]

I can't tell you for sure as I don't know what images were used to train the data base. You could try it in your prompts. From my own experience I've tried asian celebrities like Blackpink for example (Yes I'm a blink hahaha) and you can get sort of their likeness but not really. I would advise you though to be careful when using known figures. If you are using their names for the negative prompt trick then it should be fine but to generate images of celebrities you have to be very careful.

[@Javster]

Awesome video! What is the difference between INPAINTING and MASKING? they seem to do the same thing. Thank you!!!!

[@MonzonMedia]

No difference really, they do the same thing. Masking is more used in photo editing. 👍

[@MeesterGgaming]

would be nice if it whas possible to select the character face and save it. After that importing it back and let the AI use the head on everything you want

[@MonzonMedia]

Indeed that would be very useful but then again it would be better if there was the ability to train your own model. 👍

[@DigitalJediMaster]

I'm guessing this doesn't work with Stable Diffusion XL either, right? Pretty much generates the same person every time. I tried some filters on 1.5, but the results were just not getting the look I wanted.

[@MonzonMedia]

It does but it’s much easier now to get consistent faces by naming your character with a random name, sometimes 2 names will work too. Most models these days have a default look but you can change it a bit with names and nationality

[@DigitalJediMaster]

​@@MonzonMediaI actually did try experimenting with some nationalities like Nigerian and Senegalese, but it pretty much defaulted to the same woman I've been getting by just using the term "African". I also tried using names of people with vastly different ethnicities, but that still produced the same person if one of them was black. It seems it doesn't quite distinguish between the darker skin tone ethnicities. I've been using the Macro Realism filter, so maybe that's part of the problem. But it's the only one that produces the intricacy I'm looking for. Delicate Detail did the same thing. But I'll keep experimenting. I'll try just giving them names like you suggest.

[@MonzonMedia]

@@DigitalJediMaster start without any filters, just SDXL raw, as I said most filters have traits that give it a default look. Once you find something decent with SDXL raw you can use the same settings and seed and then apply a filter, or even use img2img.

[@DigitalJediMaster]

@@MonzonMedia So, interesting development. I wasn't sure if you meant two first names or two full first and last names. So I tried both along with differing nationalities and no filter. Two single-word names generated the same woman I'd been getting every time. So I switched up to two full names, same deal. So I switched up to two full names, and two nationalities. I got two heads. Okay, so I put "extra heads" in the negative prompt. Now each time I get two women who seem....very much in love. But on the bright side, they do look a little different than the woman I had been getting, so...progress?

[@Tedzik31]

Hi, love your videos, can you tell us how to avoid the top of head being cut off ?

[@MonzonMedia]

Hey Ted thank you for the kind words! it's very common with AI image generation because the data sets are usually trained with a 1:1 aspect ratio, so chances of getting multiple limbs are more likely with other aspect ratios. You can use the negative prompts section (remove image) and add words like "double head, multiple limbs, cropped head" etc. This won't prevent it but it will help lessen it. Here's a link to some common ones I use for people docs.google.com/document/d/15GsYUjHKRnP5nTkSxLQJGidiqH1GN-AVMmqPGFKoXrw/edit?usp=sharing. You can also experiment with sampling methods, Euler, DPM2 etc in the advanced options, some samplers are more prone to duplications, and cropping but it will also change the look slightly.

[@Tedzik31]

@@MonzonMedia Thank you fot fast reply and advice

[@jasonlee789]

Can you keep the clothing consistent too?

[@MonzonMedia]

It's a bit tricky but there is some workarounds. You will have to do some minor post production. By nature, ai image generation is random and controlling it isn't where it needs to be just yet. I'll be covering this very soon!

[@jasonmation]

@@MonzonMedia that's my problem at the moment, generating the exact same clothing. The face works brilliantly!

[@MonzonMedia]

It's a bit easier if you can run SD locally since there are more options available, still have to do some minor post production. However with the new canvas features on playground, we can get some consistency. Will make a video on this soon. BTW for consistent faces you can also use made up names, up to 2 names to get different looks. This method on the video was mainly for extracting characteristics.

[@TBOYTUBE]

Nice vid, but I get the same type of result without naming celebrity I just write "beautiful female, irish, 25 years old, long red hair" stuff like that and I get a type of typical woman. The perfume style is good but you basically always have the same perfect models whatever prompt you make.

[@MonzonMedia]

So this method is for taking certain characteristics of that person and just another option on creating consistent faces. Some models tend to have a default look so that when you put in the prompt ethnicity or certain traits, you can also get consistent faces. I've also found that if you use, 1-2 random names you can also get consistent faces. I'll be covering this in a video soon.

[@erakoraoghenekevwe2275]

Thanks for this video man.. I'm having an issue using playground a.i.. After generating portraits, the images merges together and I don't want that.. I just want a single portrait of the image. Can u please explain why it happens like that and exactly what to do to fix it.

[@MonzonMedia]

You're very welcome! Can't say for sure unless I know what your prompt is and settings you are using. Make sure to use the negative prompts, I left a link to a google doc in the description for some I use regularly. Also if you are going beyond 512x512 or 512x768 the tendency is for the image to produce unpleasing results. The data sets are trained on 512x512, any other aspect ratio will increase the likely hood of getting weird results. I tend to stay with 512x768 or 768x512, then upscale the image. Hope that helps!

[@erakoraoghenekevwe2275]

@@MonzonMedia Okay..I was using 1024x1024

[@MonzonMedia]

@@erakoraoghenekevwe2275 Yeah that would explain it. For people it's best to stay within the specs I previously mentioned. For landscapes or images that are not people, changing aspect ratio doesn't mess things up as much but it still can happen.

[@jkneifl15]

I don't understand what you are doing to get a more consistent character. it just looks like you are being a little more descriptive and playing around with the promps. Are you narrowing down the types of images you get when you remove the images you don't like?

[@MonzonMedia]

This is actually an old method now. I did it this way because of the way Stable Diffusion 1.5 was then. All it is doing is taking the traits of the celebrities so at least you have an idea of how you want them to look, rather then random people. Deleting the fails doesn't do anything. However nowadays it much easier to do with SDXL and custom models. You can just give a name to your character, 1-2 names and it will retain a consistent face. Most custom models now have a default look so giving it a name, ethnicity and certain features works very well. Getting consistent clothes is another story.

[@hk10218]

how to make a consistent vehicle, such as boat, car ...etc ?

[@tigtagvision]

hello new friend to your amazing channel thanks for the advice

[@MonzonMedia]

You are very welcome! Glad it helped!

[@tigtagvision]

@@MonzonMedia you are fantastic thank you

[@MonzonMedia]

My pleasure my friend! It's a weird concept but hey it works!

[@aegisgfx]

Im confused, if you wanted a female character, why not just mix to female actresses together instead of mixing 2 male actors then removing their 'maleness' in the negative prompt?

[@MonzonMedia]

Great question! For one, at least from my experience even when you mix the character in PAI one seems to be more dominant than the other making them still recognizable. Also if you use made up names + filters, you can get consistent characters as well. (I should've thrown that it the video too). Doing it this way you are just getting certain traits of the male figures that will give you the consistency of a fictitious female character. It seems counter intuitive but it works pretty well to get that consistency.

[@aegisgfx]

I just tried my method, works great. No need to do the male to female conversion using negative prompts at all.

[@MonzonMedia]

and it doesn't look like either of them and are you getting the same character multiple times without using filters?

[@aegisgfx]

@@MonzonMedia yep I sure am

[@PattyEverAfter]

Using Playground Ai: can u do videos: how to get full body in photos. and how to create action scene.

[@MonzonMedia]

Absolutely! In the mean time you can experiment with your prompt with full body, full boy wide angle and it also helps to describe what the subject is doing like walking on the side walk, standing on a cliff etc. Be as specific as possible. The problem most of the time with full body shots, the faces don't turn out as defined as you'd want it. So another solution I've been doing is taking a half body image in Canvas (or generating directly in canvas) and then outpaint the rest of the subject. That way you get a detailed face. I will be doing a video on this very soon, most likely on the main Playground AI channel so be sure to head over there too! www.youtube.com/@playgroundai Hope this helps!

[@tokshiva]

Thanks for that helpful video. Please tell us what kind of risks we'll face when we generate and use celebrity faces using playground.

[@MonzonMedia]

Hi Shiva, I'm glad the video was helpful. To be honest, I can't tell you 100% for sure but for things like personal use or even thumbnail images for example it should be fine. Where it's not ok is if you reproduce a copyrighted image and sell it as your own. I'm sure there are many other things to consider when it comes to celebrities so I would avoid doing it as much as possible.

[@davemorgan6013]

I'll have to see how good it is at recognizing Asian actors and actresses (chiefly Chinese and Thai). I suspect the system is heavily slanted towards American celebrities.

[@MonzonMedia]

Dave, in general there is a bias with any AI platform I've seen. Unless it's a well known Asian celebrity. For example, I can do images of Bruce Lee quite easily and in his likeness but If I try Lisa from Blackpink....nope just some random asian woman hahaha! 😁

[@gulnaracshudova]

Hi. May I use playground ai pictures in my profile as nft selling?

[@MonzonMedia]

Yes you own the rights but I encourage yo to read the TOS help.playgroundai.com/en/articles/6848773-how-does-copyright-work-do-i-own-the-images-i-create Hope that helps!

[@HappyTalesTime-Kids]

can we add motion/animation to these images with playground?

[@MonzonMedia]

Not in playground unfortunately. But you could bring it in something like pika labs or runway ml.

[@HappyTalesTime-Kids]

@@MonzonMedia Thank u !

[@ryanhowell4492]

That's so cool

[@MonzonMedia]

Definitely is! Follow up vid to come soon! 👍🏼

[@rogerc7960]

Perfume filter works best. 😎

[@MonzonMedia]

Yes it does! I like using that one for females. Technically you can put it 2 made up names plus a filter like perfume and get the same consistency however I haven't had any luck doing it without the filter and 2 celeb or made up names. I think because the filter is a "trained model" it helps with getting the consistency.

[@gumvue]

negativ ones, must be some templates, so we just copy past, and add view, its hard to type all time new ones

[@MonzonMedia]

There is a google doc link provided, you can just copy that and create your own. 👍

[@geniune168]

waiting for your next video bro.

[@MonzonMedia]

Coming soon bud! Busy week for me with other projects but I hope to release something by Monday. Appreciate the support!

[@WalkingDday]

What does « safety issue » mean?

[@MonzonMedia]

Initially it means that the system is filtering out NSFW images however sometimes the safety filter get's triggered for no apparent reason. It happens for a number of reasons, if it isn't something in your prompt that is triggering it, sometimes it's a setting that needs to change. I find that changing the number of images generated or changing the sampler will often get rid of that error.

[@stevenjordash2396]

Can you update this as I don't even see the 1.5 as an option and I am on pro.

[@MonzonMedia]

Unfortunately SD1.5 is no longer supported on Playground as of Feb 28th.

[@breanat1508]

There is only Stable Diffusion XL?

[@MonzonMedia]

SDXL, Playground v2, and Playground v2.5. SD1.5 was discontinued at the end of February

[@breanat1508]

@@MonzonMedia Thank you

[@CuriousMujtaba]

First

[@MonzonMedia]

Dang your quick! Dat what she sa......hahaha!

[@luizcardoso3995]

can I do this using my photos?

[@MonzonMedia]

Absolutely check it out here kzbin.info/www/bejne/jqiqgpKiZqykias

[@MonzonMedia]

BTW it's probably best to train a model based on your photos. Check out how here kzbin.info/www/bejne/pX_Em3esrZiSfbs

[@luizcardoso3995]

Thank you so much

[@MonzonMedia]

@@luizcardoso3995 Anytime my friend!