Leonardo AI Train Your Own Custom Models

Leonardo AI Train your own custom models and quite easily I might add. With that being said, I've gotten better results from Open AI's Photobooth but then again I could be missing something that's just not available yet? Either way It's a good first step and the training data sets may be useful for other types of models. I'm really loving this free online image ai generator. (at the time of recording) Take advantage while you can!
Prompts: "beautiful woman warrior wearing skyrim dragon scale armour , beautiful face, glowing eyes, epic war zone, full shot, intricate details, symmetrical, Todd Mcfarlane, Diego Dayer, Anna, Dittmann, Unreal 5, hyperrealistic, dynamic lighting, volumetric lighting, fantasy art"
Negative Prompts "ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, out of frame, ugly, extra limbs, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, mutated hands, fused fingers, too many fingers, long neck".
Check it out here leonardo.ai/
⏲Time Stamps
0:00 Read the FAQ
0:58 Trained dataset preview & Setting up images
2:08 Create a new dataset & Import images
3:47 How to use your trained model
5:11 Standard model examples
5:57 Try different schedulers
6:39 Leonardo style & Prompt magic on
7:25 Final thoughts
*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]

Hey everyone! Just a heads up there is a temporary hold on approving whitelisted registrations due to the heavy influx of users. Please checkout the community tab for details. Please be patient, it will be worth the wait!

[@MonzonMedia]

aaahhh I'll have to update that info, that was when it was still beta and pretty much a free for all. Free is only 150 tokens per day now unfortunately. That changed when the free plan came out and to be honest, I didn't think that was the right move. 250-300 at least would've been decent.

[@ATLJB86]

I have had great success with my trained models. Things to consider: The quality and clarity of the images you use, upscaling images lose detail.. You want high contrast images. Don’t use the instant prompt in your prompt, it’s already trained on your model so you will get that look regardless. Most importantly, the prompt itself. Make sure you use words like beautiful, highly detailed, octave render, studio lighting, etc.

[@MonzonMedia]

Appreciate the feedback and thanks for the suggestions however I used real images of a real person and yes they are quality images and I confirmed with the folks at Leonardo AI their training model isn't really tuned for real photos. However with some experimenting I've been able to improve the model a little bit. I've had better success generating AI images and using those to train. I cover a bit of it on my latest video. Definitely going to do a follow up video on training. 👍

[@RexCarla]

Thank you for this. What's the best instant prompt for training?

[@ATLJB86]

@@RexCarla that’s a little hard to explain but you want to train based on a theme. Exp: I made a bunch of splatter designs then trained Leo on them. I then use image to image, explaining the image, then include splatter in my prompt. Now my model has a splatter effect based on a pattern I designed. It’s good to use concepts that the ai already understands so it can interpret it correctly.

[@RexCarla]

@@ATLJB86 Thanks a lot, I'll try it, it's very interesting and exciting to train and generate images like this.

[@ATLJB86]

@@RexCarla Yes, you basically train the model on what you want to see based on what it understands. Exp: You want to make a dark scene of a cabin in the woods with volumetric lighting. You would then upload a bunch of images with a cabin in the woods that has volumetric lighting. Then use your prompts to change the style. Let’s say you want a cabin in the woods with volumetric lighting but you want it to be an oil painting. You would use your model then add oil painting to the prompt. Or say you wanted a carton style… Then you would say in the style of a comic book. You basically train what you want then prompt for variation.

[@tomstamford6837]

Can't begin to say how much I envy you folks who can do all this and create such amazing and imaginative art as well as teaching the rest of us the basics of how to try for ourselves. I would love to know, see a how to video, not only on how to generate a character, be it photo realistic or 'manga' style, to use in something like a graphic novel or video where you need to have more than one pose. 🙏✌

[@MonzonMedia]

Hey everyone! After a few more training rounds with "real photos" I don't think the training is ideal for that just yet. I'm going to try with AI generated images. Hang tight for a video to come soon!

[@mattcav6992]

Hi mate, your presentation style is very easy going. Appreciate your time putting this together. Cheers

[@MonzonMedia]

You're very welcome and thank you for the kind words!

[@goodfellas5702]

Greetings from Ireland. I only discovered your channel yesterday and I really love it. You have a great style to your teaching. There's so much going on with AI these days that my head is spinning. I'm 58 and have never been so excited by something but I'm also worried that I'll get addicted to exploring bottomless rabbit holes! To date I've been concentrating on language models because of how they can help me in my business but as a photographer I'm constantly drawn to visual AI. I'm a little confused though.......I understand what the engines like stable diffusion, dale-e and midjourney offer. Where I get confused is when I hear of tools like Leonardo, Photobooth, etc. (to compound the problem....there was one that you mentioned in a video from April that seems to have gone out of business) Would really appreciate a video that explains how these elements fit together. In the meantime, I'm going to binge watch the rest of your videos while on vacation next week. I see you did one on Luminar Neo. I'll start there as I have that tool and like most tools I use about 5% of it's capability. thanks again

[@MonzonMedia]

Hey there! I'm encouraged to hear that you are venturing into the world of AI and yes it is a deep rabbit hole! hahaha! I can see where it get's confusing and I do intend on making a video on this topic and other related questions I've seen come around. Basically sites like Leonardo, Prompt Hunt, Playground AI are all based off stable diffusion, think of it as an open source base model. One of the benefits of running it locally is that you can use stable diffusion totally free if your PC meets the system requirements which is usually dependent on your graphics card. For people that don't have the specs for whatever reason, sites like playground ai, leonardo etc offer it as a service which is priced very fair which allows users to use stable diffusion. Each platform is unique in their own way and may provide different kinds of models and features however they are all based off Stable diffusion. Midjourney offers the same service but it is not based off stable diffusion, they use their own data set. I hope that clarifies some things, if not let me know!

[@davidbackham100]

Thanks , for making this quick , easy , eagerly waiting for improved outcome like photobooth ...

[@MonzonMedia]

Most welcome my friend! I have a feeling the training isn’t tuned for real photos. I’m going to try to train with Ai images and see if it’s any better. Appreciate your support bud! 🙌🏼

[@MonzonMedia]

So I made a few interesting discoveries on training models, update to come soon! Teaser here--->www.youtube.com/@MonzonMedia/community

[@GolpokothokRaktim]

Now a days anyone can use AI and can create amazing images quickly, easily, and affordably. I recently used BlueWillow and it was super cool.

[@Grimmona]

Thank your for introducing me to Leonardo, I like to combine it with pictures from playground. The lineart in Leonardo is so much cleaner that in playground

[@MonzonMedia]

Very welcome my friend! I think it's great for us to have so much access to all these new tools! So much fun! 😁

[@MyTubeAIFlaskApp]

Very nice information on Leonardo's fine tune training. I use Leonardo a lot to make faces for speaking avatars.

[@MonzonMedia]

I appreciate that! That's awesome, what are you using to make the speaking avatars?

[@MyTubeAIFlaskApp]

@@MonzonMedia I use Python. It is my own Flask program. I am building avatars now to show how the program works and how it was was created.

[@MonzonMedia]

Oohhh very interesting! I’m hoping to learn more Python eventually. Working with SD I want to know what’s going on in the backend. I’d be interested to see what you’re working on whenever you are ready to share it. 😊

[@MyTubeAIFlaskApp]

@@MonzonMedia Keep an eye out I have several videos.

[@MonzonMedia]

@@MyTubeAIFlaskApp will do my friend! 👍🏼

[@ShortMemePlay]

Thank you for making this video. You are a good teacher.

[@MonzonMedia]

You're very welcome and glad it was helpful. Thank you for the kind words, that made my day ☺

[@solomonohonsi3214]

Wow, I have been wondering when easy training will be implemented by platforms. I hope platforms like Playground will also add this to their features.

[@MonzonMedia]

Definitely a game changer ay? I've always felt that if any platform can provide training and outpainting would really set the bar high and here we are! Very excited to see how this progresses!

[@miketrujillorealestate]

Thank you for the help, I wish you success always.

[@MonzonMedia]

Welcome my friend and hope your training goes well!

[@Rherman1950]

thanks for making this on how to make a model. Clarified it a lot.

[@MonzonMedia]

Glad it was helpful Rick! I'm not sure if it's tuned for real photos yet but I'm still testing things out. More to come soon!

[@JimArthurVanWyck]

Hello good person…another valuable video! Thanks for this and all you do

[@MonzonMedia]

Hello good Jim! haha! You are welcome my friend. I'm going to do more testing with this. I think it will produce better results training AI images. I'm not sure if it's tuned for real photos just yet?

[@spacekitt.n]

When im doing face swaps, I've found that the character setting works better than the photography setting. And 1.5 SD is the better training to use at the moment, till they add edge/depth/pose to image settings for v2 models. I've been training the same pics with 1.5 and 2.1 and both work well for different things. I've found that pics that are too lens distorted or taken at to high or low of an angle dont really work (SD models dont do well with these kinds of pics atm). And variety of lighting is a must, lighting that emphasizes head shape may be the most important, so rim lighting and shadows for some pics, but flat lighting for others (you can fake this in photoshop or use a tool like clipdrop relight if you dont have it in the original pics). I'm not sure why this works but through lots of wasted tokens and trial and error i've come up with something that kind of works for what i need lmao. Thanks for all the info in this video though it gave me a jumping off point and led me down a rabbit hole that has helped me really grasp what SD is doing with custom models lol. Powerful stuff but its a lot of work and specific steps to get there

[@Itsrichayo]

Great content. Thank you for sharing.

[@MonzonMedia]

My pleasure! I need to do a follow up video on this! Hopefully sooner than later 😊👍🏼

[@nskrynski]

This was an excellent tutorial, thanks

[@MonzonMedia]

Thanks Noel, I appreciate it! Be sure to check out my follow up on this video, about halfway through. kzbin.info/www/bejne/r6TKYWmlgax0b5Y 👍

[@ed2win37]

Thank you so much for this video. You are my new hero!!!!🥰

[@MonzonMedia]

You are very welcome! Glad it was helpful 😊🙌🏼

[@rodron1]

Hey Ermin, long time no see, hope all is well. Great videos man, I've been playing around with Leonardo ai for some time as well. I'm liking it a little more than Dall-E, although I've yet to play around with MidJourney. I've subscribed to your channel, this is great stuff.

[@MonzonMedia]

Hey Rodney! So great to hear from you and I appreciate the support! I had no idea you were into AI as well! That's awesome! So this was one of my earlier videos on Leo so as a user I'm sure you know there have been a few more features added since. I will be do a follow up video very soon on those things. Hit me up anytime you want to chat or if I can help out in any way.

[@EduardoTorresDeMadera]

This was an excellent video. Keep up the good work. I do have a question. One I create the model, will this model be available to others or just for my own use? Thanks.

[@MonzonMedia]

Thank you! At the time of recording there was an option to keep the model private or public. For free users it may be different now though as things have changed since I made this video. You might want to check with their Discord support or email them to be sure. 👍

[@johnleighdesigns]

Great to see this one - seems im catchuing up since only recently started my journey with Leonardo which seems to be my favourite AI tool so far - I do like also Playground but something about Leonardo keeps me coming back testing, experimenting! One question if I may get your wisdom i tried this Training and results are not really great but did find that the image to prompt optioni combined with a trained model works quite nice after some fine tweaking -

[@MonzonMedia]

Did you use real photos or AI generated images? From my own experience, it's really not the best using real photos. At the time of recording this video the folks at Leo did confirm that things would need to be tweaked for that. Not sure if it's happened yet. It will take some trial and error, if you are training people it's best to have at least 3-5 images that are full body and half body, different backgrounds and angles. The rest can be close ups of the face in different angles and backgrounds. It does look better when you apply any of their upscaling methods. 20-25 images is usually enough but more could work better but it can also over fit the model. If you watch this video, I do a follow up kzbin.info/www/bejne/r6TKYWmlgax0b5Ysi=Ej37t2edOu2492mT&t=251

[@johnleighdesigns]

@@MonzonMedia yes used real photos,and some of the renders do work but only if add image as prompt too. I shall explore further but it eats all my free tokens, seems it's time to become a paid subscriber

[@seoathlet2199]

Thank you :))

[@smuki196]

Hi, thank you for this video. One question, what is the tool used for the retouching at 5.17?

[@MonzonMedia]

It was a combination of Photoshop and Luminar Neo as a plugin, however look at this follow up video where I show how the upscaling helped improve these images. kzbin.info/www/bejne/r6TKYWmlgax0b5Y

[@wizzNarts]

Thanks for the inside... What would be the better way when training a model? a) different images of a model in more or less the same pose or different images of a model in all sort of poses?

[@MonzonMedia]

Depends on what it is, if it's a person/character, you want to have different perspectives and angles of the person. some close up, half body, full body, different clothing etc. Good to mix it up. I will say though that for people, I think they need to fine tune the training more and I've been in touch with them with my findings and they will provide feedback soon.

[@wizzNarts]

@@MonzonMedia Thanks very much - I think it is a process of learning for both the dev-team and the machine. I kinda like the output for what I am doing at the moment which is writing books. The images help visualizing the characters. But I would love to see a function that "freezes"/"memorizes" marked areas so that resampling keeps those things, lets say, the clothes. For now I will bend the canvas features as far as I can... :) again, thank you very much for your patience...

[@Lunangel]

Really helpful! Thank you so much. I have a character I created but I only have around 13 images of her. Would these suffice to train an accurate Leonardo model?

[@MonzonMedia]

Minimum would be 15 but for people you would want more. It’s kind of hard to say and will take some experimenting. Personally I use 20-30 images for people. You want to use at least 3-5 close ups of the face and 3-5 full body images and the rest can be half body or 3/4 length in various poses and backgrounds.

[@Avalon19511]

Here are a couple of questions one of which has prevented me from doing any training, first, local training I'm told takes at least 12 gigs (I have 8) so is the process done through their servers, and second can we use that model on our local running SD?

[@MonzonMedia]

Hey Avalon, so the training is done on their servers so GPU specs don't matter since we are not running it locally. The downside is they do not provide a CKPT file that we can use locally. However I suspect that will become available in the future. I'm going to ask about this too and I'll keep you all informed.

[@JoKeR-hl1np]

thanks for the great video AND GREAT explanation, did the output model is full ckpt model or it's lora model? hope they add ability to download model and make it as LoRa, this will be better, did the model i train will be private?

[@MonzonMedia]

I can't say for sure unfortunately but I can ask. There is an option to keep models private.

[@jestercorner]

Thanks for this video! Unfortunately, when I try to start training the model, an error pops up: An error occured submitting your request: internal error. Can you suggest what is the problem?

[@MonzonMedia]

It could be a number of things, their own server could be the issue. The other thing is the data used. Did you make sure to use 1:1 images 512x512? Also what plan are you on? Free or paid? Certain tiers only allow a certain amount of training per month.

[@hemaldecosta8220]

Hi, In the Leonardo FAQ, it is advice to upload high-quality images when creating a Custom Data set. The highest supported dimensions currently are 768 x 768 for training. Would it make a difference in the output quality of the generated output if we upload images in higher resolution than 768x768 (keeping the aspect ratio)? I just uploaded 768*768 images for the dataset and the faces in the output from the created model are not as good as the results from the available fine-tuned ones. That's why I'm wondering whether it might need some high-res images to train properly.

[@MonzonMedia]

Hi Hemal, so a few things. If you are using real photos, your initial generation may not look that great, but try using the creative or alternative upscale. It will make a big difference. I also confirmed that the training in Leonardo AI isn't necessarily tuned for real photos but they will make improvements to it soon. If you were using AI generated images, it's likely you would get better results from training. If you used 768x768 that is only for SD2.1. Although the training can be done, it's best to use SD1.5 and train at 512x512. It's just more versatile. It's always best to use images that are at its highest resolution even though you are cropping them to the given specs. Hope that helps!

[@MonzonMedia]

BTW check out this video where I go over how upscaling affects the image, and now it's even better with 2 upscale options. kzbin.info/www/bejne/r6TKYWmlgax0b5Y

[@hemaldecosta8220]

​@@MonzonMedia Thank you for the advice! Will try with your suggestions....

[@melanieb115]

I love this tutorial. Out of curiosity, can this be applied with non facial models? ex. different types of buildings, cars, etc.

[@MonzonMedia]

Yes, absolutely! You can pretty much train on anything but it will take a few tries to get it right depending on the subject. Just experiment. 👍

[@tenzintsenpey5274]

when I checked the job status, strength shows medium. is that anything to do with outcome of the image? I mean what if its on high mode and give better result?

[@MonzonMedia]

I noticed that as well and thought the same thing however if you look at the platform models, they all have that medium strength designation.

[@malarkey102]

Hi sorry to go of script but are you making a video on this Pinokio Ai Browser. Would like to know do I still need a good graphics card. Or does it work differently. Also do you ad models the same way as an actual installation with automatic 1111.

[@MonzonMedia]

Hey there. I thought about it as I think it makes the installation process so much easier. From what I gather yes you still need a decent system as it still runs locally. I think the main point of this is to make the installation process easier and for advanced users you can make scripts, use api etc. would you be interested in a video about it?

[@malarkey102]

@@MonzonMedia I think a video would be great making it easy is the best way

[@goodfellas5702]

Thanks

[@MonzonMedia]

Wow, thank you so much! So kind of you @Good Fellas! 🙌☺

[@poakenfold5954]

Hello! Would you know answer to this quick question? What is the quality of Leonardo Stable diffusion Training model as compared to a Stable diffusion model created on Google Colab (ckpt file) using same set of images? THX

[@MonzonMedia]

Unfortunately I can't answer that as I've not trained on collab and not a fan of using it. However I can tell you that Leonardo is limited on how they have set up their training. As I mention in the video, I got better results using real photos from Open AI's photobooth than I did with training in Leo.

[@poakenfold5954]

@@MonzonMedia Thanks, I think so too as Leonardo took no less than 10 minutes to train about 40 images which they say takes about 30-40 min on Nvidia Tesla GPU on colab.

[@jeffreyzonenshine]

Excellent tutorial. I do need help with model training. My goal is to train a model such that the generated images will contain an accurate representation of the face of the person I used as the subject. So far, my attempts have resulted in a generic face. Any advice? Would using multiple copies of a single image of the subject for the training work? Thank you.

[@MonzonMedia]

Thank you Jeffrey. I appreciate you checking it out. Are you using real photos or AI generated? Also are you using the instance prompt?

[@jeffreyzonenshine]

@@MonzonMedia Thanks for the quick reply. I am using real photos at 768 x 768. The Base is v2.1 and I am using the instance prompt.

[@MonzonMedia]

@@jeffreyzonenshine Ah ok, so a couple things. So training with real photos in general in Leo isn't optimized for that just yet but I do know they are working on it. Also, you will get better results with SD V1.5. So train your model with SD1.5 at 512x512. You also want to use a lot of close up portrait style photos in different angles but also include half body and full body shots, maybe 3-5 of each and more on the upclose images to your discretion. It might take a few tries to dial it in. Also be weary that the initial generation of the image might not look great but when you upscale the image with the "alternative upscale or creative upscale" it will give better results. Check out this video where I share some of my findings when using real photos vs ai generated images. kzbin.info/www/bejne/r6TKYWmlgax0b5Y Hope that helps!

[@jeffreyzonenshine]

@@MonzonMedia Your advice was helpful. I did get better results with SD V1.5. at 512x512. However, the results are quite random. Using the exact same inputs I will sometimes get results similar to the original results but more often I get the equivalent of graphic gibberish. In short, as you indicated, the tool does not seem ready for use with photos. I will stay tuned.

[@imyashw]

thank you

[@MonzonMedia]

You're welcome!

[@accentedreality]

Have you tried using the newer canvas features to fix the eyes, hands, etc?

[@MonzonMedia]

Yes if course, I cover that in another canvas video but this is a fairly old video now.

[@shlomitgueta]

After I uploaded photos and trained the model and then decided to add photos or download packages, do I need to click on training again? If so, it doesn't allow me more than twice a month

[@MonzonMedia]

Yes if you want to add photos and retrain you will have to train again. Depending on your tier you can get more training but for the free account it's limited to 2. Hope that helps!

[@dirkbaeten6949]

I used 9 out of my 10 trainings for this month. And I am still trying to find a better way to be able to get a character consistently showing in different poses with the same outfit and hairstyle. When you only have a picture from AI art, it means it's very hard to make multiple pictures with the same outfit/hairstyle etc. showing different sides of the same character. So I just uploaded multiple copies of the same picture and it kind of works, but still to get the same outfit with the correct colors is a big challenge. I wonder if you have any advise for that, or if you could do a tutorial for that.

[@MonzonMedia]

I get this question a lot and the short answer is, AI generated art is totally random, so no matter what you do, there will be slight variances. There are work arounds to come close but not exactly. With that being said, I've been working on a method behind the scenes to get close to what we want to see in terms of consistency. Just need to test more before I share my findings. Also there will most likely have to be some post processing regardless to get that consistency. This is true for any AI platform, even Midjourney.

[@dirkbaeten6949]

@@MonzonMedia Ok. Thanks a lot for the reply! Good luck with your new method.

[@MonzonMedia]

@@dirkbaeten6949 appreciate you! Will share my results soon!

[@MonzonMedia]

just a little teaser www.youtube.com/@MonzonMedia/community

[@spacekitt.n]

@@MonzonMedia yes the best possible scenario for getting your trained images I would imagine is to use either a real person and photograph them with studio lighting with diff lights and angles--or if its not a person, to use a 3d modeling program to create your character and pose/light them in different ways in the 3d software.

[@cutefairykiller]

What is the point of adding the prompt if you eventually select the model itself ? Is it meant to reinforce the training ?

[@MonzonMedia]

Are you talking about the instance prompt? So generally speaking it depends if you are training for a person or style. If it's a style the model will pick it up regardless if you enter the instance prompt. Adding the instance prompt will just further reinforce the image to give you exactly that. If it's a person, the instance prompt is needed to look like the person you based the model from. Otherwise the faces will just be random.

[@asadjamil6127]

One Problem I'm currently facing is that, it duplicates key object in the image. For example, i would write watercolor style cute butterfly and there would be two messed up butterflies in the images. How can i resolve this issue.

[@MonzonMedia]

Try adding in the negative prompt "duplicate, multiple" words like that. you may have to also use inpainting to correct those things. Take a lot of trial and error.

[@pvgmartin]

Can you download the trained model checkpoint from Leonardo for use in other SD platforms wile Automatic 1111 or Comfy UI?

[@MonzonMedia]

Unfortunately not at the moment.

[@AlexGNewMediaJournalism]

Does the models we create in Leonardo available to everyone? Or is there a way to make it a private model? Thanks

[@MonzonMedia]

Hey Alex, as far as I know they are private by default with an option to share it with the community. But I’m not 100% sure. Best to check with them either on discord or support.

[@AlexGNewMediaJournalism]

@@MonzonMedia i thought so. I haven't tried it yet although im already on the early access..great video, thanks.

[@SpectacularSpidey670]

awesome! the only problem is that you can only upload 40 images, for me thats the problem, is there other AI that does the same but that can upload more then 40 images? i dont have acess to stable difussion

[@MonzonMedia]

No need for more than 40, usually anywhere from 15-25 is ideal, 30 at the most.

[@SpectacularSpidey670]

@@MonzonMedia its because i want to replicate an art style, so that requires more then 40 images

[@ShowbizShortsz]

Maybe a silly question here but can you bring your Leonardo model to say tensor?

[@MonzonMedia]

Not silly at all but unfortunately they don’t let you down load the models you train

[@Hanna_Kulazhenko]

Hey) is "step count" function available only if i upgrade coins? Can't found it...

[@MonzonMedia]

If you are on the free version you are limited to the "leonardo scheduler" which doesn't have a step count but is usually more than enough for most. Paid users can use other schedulers with step count.

[@Hanna_Kulazhenko]

@@MonzonMedia thanks)

[@juliagarcia7302]

does anyone have a tip for training a model on obejcts with text on it? the text seems to allways be very "hyroglyphic"....

[@paulsheriff]

do you have a link to PHOTBOOTH or a video on it ..or one on how to load stable defusion locally cheers and good video

[@MonzonMedia]

Check these out! How to install Stable Diffusion V1.5 kzbin.info/www/bejne/r5S0e3eAYtaBf5o (courtesy of @sedetweiler ) How to install Automatic1111 kzbin.info/www/bejne/aZTZgWqvf9Sni68 How to install F222 Model kzbin.info/www/bejne/e4uWoKmFnq-ghtE (courtesy of @OlivioSarikas ) For photobooth go here openart.ai/photobooth

[@edtomlinson1833]

Why do you have the Canva step? Why can't you directly upload the (jpeg) images from your files?

[@MonzonMedia]

Images need to be cropped to a 1:1 ratio. Any editor will do. But at the time of recording this video this was based on sd1.5

[@regisvoiclair]

Thanks, but how to pass from "BEFORE" to "AFTER" at 5:12 please ?

[@user-cr6lc9dm4y]

That's great, thank you! But the AI changes the clothing in each image. How can I make it stay the same clothing ? Thank you!"

[@MonzonMedia]

Hey there, yes it can be tricky to get consistent clothing, by nature stable diffusion is random. I do go over some workarounds in this video kzbin.info/www/bejne/f4GtgoJrqNOAmpY and also you can prompt as you normally do but increase the size to purposefully get 2 or more people in the image which you can then in post, split them up.

[@ericr2617]

Training starts says "done" but my model still doesn't appear in "your models" ... Any ideas??

[@jocelyn-n-tech]

Please do a video on how to fix the eyes in post.

[@MonzonMedia]

Absolutely! Already on my list of topics to cover. ☺👍 Appreciate the feedback!

[@yusrileka1036]

So more picture reff will make greater trained model?

[@MonzonMedia]

Not necessarily Yusril, there is such a thing as over fitting a model where too much can make it worse. Finding a sweet spot can be a challenge sometimes. Highest resolution possible is key and depending on what your subject is, you may need variety of angles and perspectives when it comes to people.

[@MassimoPardini]

CAN I TRAIN FOR SEASCAPE USUALLY DIFFERENT WAVES BACKGROUND OR WILL MESS UP?

[@MonzonMedia]

You can train styles, scenery I'm not sure....I would think it will be pretty random though.

[@anjasnugraha9350]

Thank for sharing. But, why i don't have access " Start Training " . Can you help me?

[@anjasnugraha9350]

Btw, i use this free.

[@MonzonMedia]

Have you already trained a model? There is a limit to train for the free membership which is 1 model a month.

[@anjasnugraha9350]

@@MonzonMedia ahhh i see. Thankyouu

[@dangyeonsim828]

I tried to edit my database and training the model but it keep saying "Out of Model" so i can't train it. What should i do now?

[@MonzonMedia]

Again, you are only allowed one training per month. Did you do the last model this month?

[@dangyeonsim828]

@@MonzonMedia Yes. So you mean it will reset automatically every month?

[@dangyeonsim828]

@@AquiktalSubs It reset now, but i haven't tried it yet. I will when i have time.

[@paxwebb]

How long before an invite is likely to get sent after signing up for Leonardo AI?

[@MonzonMedia]

Hey Pax, typically it's within 24hrs. However they are still experiencing a surge in new users so they've been trying to just keep up. I'd encourage you to check the discord for latest info. How long have you been waiting?

[@paxwebb]

@@MonzonMedia Yeah, I hadn't introduced myself in Discord or filled in their questionnaire that's apparently required to sign up. So I did that, but now invites are on hold 🤣

[@paulb88916]

Whatever happened to unzooming? I can’t find it anywhere now.

[@Fclefland]

So basically the face is the important one? Or you need to train the model with the body included?

[@MonzonMedia]

Hello Fclefland, Typically if you are training for a specific person then yes, best to have more portrait type shots at least 3-5 full body and 3-5 half body. various angles, backgrounds and outfits help as well. It can be a bit tricky because there is a sweet spot that is unique to your training model so it might take a few attempts. One thing to keep in mind if you are using real photos, like in my video, the results are typically better after using one of the upscale methods. If they are AI generated images, then upscaling isn't absolutely necessary. The current training with Leo still needs some adjustments for real photos. Be sure to check my findings and updates here kzbin.info/www/bejne/r6TKYWmlgax0b5Y

[@Fclefland]

@@MonzonMedia your video helps a lot.. thanks so much sir. Subscribe 😊

[@dwainmorris7854]

I see that you used the AI to train your custom face but you use a costume from the AI database . but how do you get the AI to use your costume design that you created

[@MonzonMedia]

Hey Dwain! Thanks for checking it out and for the feedback. Actually the outfit or armor is in the prompt. Check the description. It's based on Skyrim armor sets. It's not exactly like it but it does give the armor a unique look. Try it out, so fun! 👍

[@emersoncollazo1875]

I did trained a model with the face of a person but lately I have observed that all images I get from using this model the face gets distorted, something that did not happened with the first images. Does anybody experienced something similar?

[@MonzonMedia]

What model are you using? Could be that but also upscaling will help fix those images sometimes, depends how severe it is.

[@donschannel9310]

before watching your video i deleted something in my model and im new to that site ive been looking for how to add my model but somehow i deleted something and it doesnt show my model face :(

[@donschannel9310]

i deleted the tab in my model because it wasnt showing the face of my model

[@MonzonMedia]

Best to contact their support on discord or email. can't help you there without seeing the issue myself.

[@happierhealing]

Anyway to use multiple models in the same image?

[@MonzonMedia]

Not that I’m aware of although you could probably generate separate images using different models then blend them in canvas.

[@Drathaar01]

So... how do we fix hands and eyes in post?

[@MonzonMedia]

Depends on the severity I’ll be covering this soon!

[@milestrombley1466]

Can you add models from Civitai?

[@MonzonMedia]

Unfortunately not but they do have models from there. You can request it as long as it licensing uses allow it.

[@iloveqq2011]

What are the purposes of training our own models?

[@MonzonMedia]

For one, consistent characters, well at least faces. Also if there is a style that you want consistency on for things like character development, comic illustrations etc.

[@brainiac2417]

How many models can you make using the free plan?

[@MonzonMedia]

I believe it's only 1 model per month and you can only retain 1 at a time.

[@brainiac2417]

@@MonzonMedia Thanks for your quick and informative answer. Great vids - keep it up!

[@dangyeonsim828]

It's not working, training database keep saying "Out Of Models" and the button "Training Model" is blurred, there's nothing i can do now with training my own model

[@MonzonMedia]

If you are in the free plan you are only allotted 1 model. You would have to delete the existing one and train again. I believe it’s 1 model/mth but double check with their support.

[@dangyeonsim828]

@@MonzonMedia I deleted it and it still says "Out of models" so it mean I can't create a new one again either.

[@hollywood_icons9181]

I dont know what I'm doing wrong , but I cant find my pictures I have uploaded, appear this category but without pic, appear like fully empty , anyone knows what I could be doing wrong ?

[@MonzonMedia]

You mean for the images to train for the data set?

[@juanreal1152]

@@MonzonMedia yes. something Im doing wrong. I uploaded them but there are 0 pics in my date set

[@MonzonMedia]

@@juanreal1152 make sure they are 512x512 or 768x768 for SD2.1

[@juanreal1152]

@@MonzonMedia I will try. Thanks mate

[@MonzonMedia]

@@juanreal1152 keep me posted

[@nivukachishi5206]

It says internal error when i press on generate. What should i do?

[@MonzonMedia]

They were doing some maintenance on their servers not long ago. Should be done by now? Join the discord to keep up to date with the status of the site. They are still experiencing and influx of users.

[@nivukachishi5206]

@@MonzonMedia Thank you so much

[@kmlchnl]

can we download the ckpt file after training?

[@MonzonMedia]

Unfortunately not at the moment. I suspect it may be one available in the future but I don’t know at this time.

[@user-wq8yq9el6v]

I've run out of models, it doesn't work, why isn't it available I haven't been able to train anything for 3 months

[@MonzonMedia]

They changed it when pricing was announced I believe free tier can only train one model. Best to check with their support by email or discord

[@miketrujillorealestate]

How can i train a photo of me in Leonardo please help?

[@MonzonMedia]

You will need various pictures of yourself in high resolution but cropped to 512x512. Also minimum 3 full body and half body images, the rest can be portrait style. Variation of backgrounds help. However unless there has been an update to their training, it doesn't really do well with real photos as you saw on the video. It's acceptable but not the best result. Make sure to check the follow up video on the results I got when upscaling here kzbin.info/www/bejne/r6TKYWmlgax0b5Y

[@miketrujillorealestate]

@@MonzonMedia Thank you for the help, I wish you success always.

[@RiseUp.88]

"How many models can be trained on Meet Leonardo AI?" Thank you

[@MonzonMedia]

Depends on your plan, free is only 1/mth, then depending on the subscription tier it goes up from 10-50.

[@Fclefland]

If the model using glasses are the glasses stay in face?

[@MonzonMedia]

If possible remove them, glasses usually don't do well with training. 👍

[@Fclefland]

@@MonzonMedia thanks so much 🙏

[@MonzonMedia]

@@Fclefland welcome!

[@faceshredder2576]

does anyone know if you can delete the photos that the AI have already been trained on and put new ones in to train it with the new photos?

[@MonzonMedia]

Hey there, you can only do that with models that you've trained, not the community or platform models.

[@faceshredder2576]

@@MonzonMedia oh no I understand that! I was training a model for stickers and was curious if I can remove the photos that I have imported in to train it. My question is if I can delete those original images, would it retain its training with those original images if I put in new ones? Thanks for taking the time to answer! Great video! 👏🏻

[@MonzonMedia]

@@faceshredder2576 AH! Of course hahaha! So yes, you can just go to the left menu under training and data sets. There you will find all your trained models. Simply hover over it and you will see an edit dataset option. You can then delete, change, swap images as you see fit. Just select the images you want to delete and you will see an option to remove images. You can also upload different ones as well. I've done this numerous times to dial in the models I'm testing. Hope that helps and sorry for the confusion. Let me know if you have any other questions.

[@faceshredder2576]

@@MonzonMedia to be extra clear I can go in and delete the original images and the AI will retain the training knowledge? I finally got my model to give me different ethnicities correctly so I’m very excited about it! I spent two days on this and I’m finally getting to where I want the AI to be 😅 so sorry if I may be annoying I’d be devistated if I lost this training haha

[@MonzonMedia]

@@faceshredder2576 No worries, better to be safe than sorry. So when you re-train you are making a "revision" which will in turn create a new model. So in that case I would just create a new data set and train a new model with your revised images. The AI doesn't work that way, it's already trained and basically we are giving "input" on what to train the model on. It does learn about the images we input however it's not like it can retain, previous information and learn from that point. Hope that makes sense?

[@alexxcabanas]

has been a nightmare for me the rezise on canva, my images are 1500x800 , can anyone give me a solution please? if i try to rezise, they stretch horrible

[@MonzonMedia]

For training they have to be in a 1:1 ratio 512x512 so you will have to crop your images. What are you training? A character, style, object?

[@alexxcabanas]

@@MonzonMedia myself hehe, I did crop, put them at 512 and they're super weird, I think I need a better camera :/.. Thx for the video and the tip bro

[@MonzonMedia]

Always best to start with good images but it also might just be you need to adjust the type of images. If it's images of yourself, try 20-25 images with most of them being portrait style in different angles and backgrounds, have at least 3-5 images of full and half body. It may take a few tries to dial it in. With that being said, I've learned that as of now, Leo's training isn't really optimized for real photos of people. Although after upscaling with creative and alternative upscalers, I've gotten decent results. Check out my follow up here. kzbin.info/www/bejne/r6TKYWmlgax0b5Y I've been told that they will be adjusting the training model to accommodate using real photos of people soon.

[@nitsujxeon]

how to fix moderation filter error?

[@MonzonMedia]

Can't help you with that bud, check with their support by email or discord. That's an internal thing that I have no control over.

[@SpectacularSpidey670]

dude answer me, is there other ai that upload more then 40 images, i want to replicate an art style, you dont understand, i want to create new things, thats why i need more then 40 images

[@MonzonMedia]

The only online option I can think of is Open AI's Photobooth but it's a paid service, you can upload up to 100 images for $30, 6 models. openart.ai/photobooth/packages?option=normal. Other than that the only other option is running automatic1111 and training with dreambooth but requires a hefty 24gig vram GPU. Google collab is another option but that's beyond me, you'd have to rent GPU time through google or a service like runpod.

[@kozavr]

The biggest problem with Leonardo is that you can only train one model on free plan. Not "have" but train. Even if you remove the model, you can't train more models.

[@MonzonMedia]

Yah that’s unfortunate, at the time of recording this video there wasn’t a paid plan and it was beta. Them when paid plans came out they restricted it to 1model total but I guess I can understand as it’s not cheap to have to have servers for both image generation and training models.

[@reno5265]

Does anyone know what he meant when he said about. photobooth model?

[@MonzonMedia]

It's a service where you can train models, starts at $10 but the benefit is that you can download the ckpt file and run the model locally on automatic1111, Invoke AI or any other local SD platform. Check it out here openart.ai/photobooth. Hope that helps!

[@reno5265]

@@MonzonMedia Thank you! Very helpful! So I can then make my photos (for example) in any style with any prompts, not just from presets?

[@MonzonMedia]

So if you purchase one of the packages, you can generate images on the site but you are only allotted a few credits and I think limited to 512x512. Again if you can run Stable diffusion locally with something like Automatic1111 or Invoke AI then you are free to create anything with that model you can use it to create the same character in many ways, or use it as a "style".

[@bura9448]

broo plz make a video on how wee made our own toddler type photo like as midjourney on leonardo ai ❣️❣️❣️❣️ love from Qatar

[@MonzonMedia]

Hello! Thanks for watching from Qatar! Amazing! I can probably add it to a future video. If you know how to do it in MJ it will most likely work on Leonardo AI too. 👍

[@GoldKingsMan]

New to channel.

[@MonzonMedia]

Great to have ya! I appreciate the support! 🙌👍

[@AI_Creator_IA]

Great Video. Thank You hand-purple-blue-peace

[@MonzonMedia]

You're very welcome!

[@AI_Creator_IA]

@@MonzonMedia Thank you! You welcome too ;---)

[@jserenity]

I cant train dataset now. it is saying out of models

[@MonzonMedia]

Unfortunately they have changed the limits for free users. You might want to confirm with them directly either on discord or by email. I heard that you are only allowed to train one model now but I don't know if that means 1 model at a time or 1 model lifetime for free users.

[@jserenity]

@@MonzonMedia , I did my test train as I was still trying to learn and waited for a month cause it says to wait for a month to be able to do it again. And now they totally restricted it? I was ready to buy the credit after I test if it really gonna take the pic input I put. Too bad then

[@MonzonMedia]

@@jserenity when I recorded this video there were no paid plans out and not as much restrictions. A lot has changed since then. Unfortunately because I work for playground now I’m not keeping up with Leo since I can no longer cover them. I’m any case I hope it works out for you. 👍🏼

[@SamuraiG]

You give some great info on ai things but for Leonardo you can only train now if you are a paid subscriber, you used to be able to train I think 4 as free but they took it away. Leonardo is taking things away from the free accounts and not giving them things like multi select to delete this is also paid only. I don't know about anyone else but I would like to clean up my feed but its hundreds of images and im just not going to do it 1 by 1. Multi select should have been on the platform from the beginning its a basic ability of image browsers. Leonardo is still an awesome generator and despite these things I am not going to quit. Well until I get a new PC and can do automatic 1111, looking forward to learning lots of new things.

[@MonzonMedia]

Hey there! appreciate the feedback. At the time of recording this video, Leo was still fairly new and there wasn't a paid plan at the time. It was something like 250 tokens after the initial release which was much more. But since paid plans were introduced tokens became limited for free users. Even training was more at one point but now it's only one model allotted and if you want to train a new one, you have to delete the existing one, unless you are a subscriber. I do like that they are really focusing on image quality with the recent features but those tokens add up very quickly. There are other options out there, have you tried playgroundai.com? Although there isn't any training you get 1K/day free. I should disclose I do work for them now...not the reason why I'm telling you though hahaha! Just giving you other options.

[@SamuraiG]

@@MonzonMedia yeah sorry did not see the age, don't get me wrong Leonardo is awesome probably the best since midjiurney has no free plan anymore. The 25 gens were very very good but I like Leonardo better and yes though playground has more gens and other things it's still not quite up to Leonardo. Sorry to your employers. 😊

[@ExodusPrime1]

Do you have to pay to do this??

[@MonzonMedia]

They've changed the usage for this since the time I recorded the video. Free members can only train one model. Anything more you have to buy credits.

[@yukoyamamoto6457]

Wish to see Train Model for Ryza

[@MonzonMedia]

Unfortunately I won’t be able to show it my friend. I work with Playground AI now so it would be a conflict of interest. However I am creating a new one of her to run locally. I might share that in a future video

[@yukoyamamoto6457]

@@MonzonMedia I understand. Because I am still new at this. Still can't grasp what it does. All we need are pictures? And then the software will edit it? Edit the whole body to make different/attractive etc.?

[@MonzonMedia]

@@yukoyamamoto6457 This particular function is for training a specific style or look. It will create a model that you can use to generate the same style or character. In my case it's of the same person. So yes, with this model I can create any scene any look with this particular model and it will always look like April. The nature of stable diffusion is to create random images, getting consistency is one of it's biggest issues. There are some work arounds to get consistent faces via prompt however training is more of an advanced way to train the model to look like a specific person. Some also use it to train a particular style whether it be based on an artist style or a style that the user has come up with.

[@visioncreadora3633]

Sorry but, what is Photobooth?

[@MonzonMedia]

No need to be sorry 👍🏼 it’s basically an avatar generator openart.ai/photobooth very reasonably priced but the biggest advantage is I can get the model file and run it locally. 🙌🏼 hope that helps!

[@visioncreadora3633]

@@MonzonMedia thank you!!! great work!

[@MonzonMedia]

@@visioncreadora3633 you’re welcome! Appreciate you! 👍🏼

[@01aaunkhan64]

My processing failed why?

[@MonzonMedia]

Not sure if I don’t have context but you’re better off check with their support

[@01aaunkhan64]

@@MonzonMedia solved thanks for reply

[@MonzonMedia]

Great! 🙌🏼

[@souravbijlwan5029]

Make some avtar videos on Leonardo AI

[@MonzonMedia]

Sure no problem however I did do one for playground ai, same concept applies. kzbin.info/www/bejne/jqiqgpKiZqykias

[@video_photo_editing]

I can't get photos that look like my trained models

[@MonzonMedia]

When you trained the model you were to input an instance prompt, are you using that word in your prompt? Check 3:10 of the video.

[@video_photo_editing]

​@@MonzonMedia yes used

[@video_photo_editing]

​@@MonzonMedia maybe when the model was trained, the category of the model had to be selected "all" and not "photo"

[@MonzonMedia]

@@video_photo_editing hhhmmmm no that shouldn't affect it. So there is such as thing as overfitting the model, like using too many photos, too many variations or even the opposite. You might have to play around with it more. What is it you are training? A person, character or a style?

[@video_photo_editing]

@@MonzonMedia I trained myself, the man wanted me to be dressed in iron armor like you have in the video, used 20 images 1.5 and 2.1 tried to train so it doesn't work so that my face in the picture comes out 😒

[@ryanhowell4492]

I want to try

[@MonzonMedia]

Head over to leaonardo.ai and sign up, the approval process is usually within 24hrs however they've experienced and influx of new users so it could take longer.

[@ryanhowell4492]

​​@@MonzonMedia frustrating part is having to wait 24 hours for my credits to be restored

[@MonzonMedia]

@@ryanhowell4492 Yes I know! Are you on the free or paid plan?

[@ryanhowell4492]

​@@MonzonMedia I'm on the free plan

[@Ashleyapples]

I hope I get an invitation

[@MonzonMedia]

They tend to send the invites in batches, it will come! 👍

[@Ashleyapples]

@@MonzonMedia i just got invite, and i trained, but the images looked distorted blob of some sort

[@khalilbenalgia4279]

you cant no more , you need to pay

[@johncressmanci]

Ugh! They aren't whitelisting new people at the moment!

[@MonzonMedia]

I just posted a notice on the community tab to give you all a heads up. Unfortunately they are still struggling with being over capacity so it's just a means of controlling users. I'd go ahead and register and be patient to receive the email on further instructions. It will be worth the wait!

[@ceesseetherdet8663]

Playground AI is never available for creating images.

[@MonzonMedia]

What kind of issues are you having?