FLUX + LORA and Kling AI (Consistent Characters & AI Videos with Your Face)

[@AskJoe]

This is the best use of AI I've ever seen... Mona Lisa was immediately recognizable and relatable as a real person you could have a real conversation with.

[@belokol]

I'm agree with you

[@carstenhankel6287]

crazy

[@effektgeraeteinfo]

I like the 6 finger hand at 0:24. Total crazy! 🤣

[@Марго-ц4о]

Настоящая Мона Лиза! Оч круто! Поздравляю!

[@PopGlide7]

The most challenging part of making videos for me is maintaining the consistency of the character's face, but this seems like the best method.

[@chiseen]

is this the same case if it was just a specific part of the body? lets say hands? a 1 hour video of streaming video of consistent looking hands doing hand model movements?

[@PopGlide7]

@@chiseen If you want to make a video with consistent-looking hands, it might be better to create a LoRA model of the hands you want and then generate the images using that model before making the video. That could help maintain consistency.

[@chiseen]

@@PopGlide7 awesome. I'll look into it. I appreciate you responding, it's really great help

[@mehmetalirende]

A question @cyberJungle: Lets say you have Mona lisa while playin tennis in that picture, but what is the method to do : multiple pictures of mona lisa while playing tennis, with same clothes, same sun shadow, same racket, same makeup, same clothes, same enviroment, but diffrent poses.

[@thewebstylist]

WOW now these are most realistic I’ve seen!

[@DAVYNIMAL]

Incredible, yes 😧

[@altairvianna8749]

Muito parecida , aliás os dois, que privilégio a natureza nos dá ❤a oportunidades de sermos parecidos com alguém , uma raridade divina atemporal❤

[@wallaguest1]

the fire light is absolutly spot on, like it understand the environment with perfection, is just incredible.

[@benbbritt]

Do you have to make separate models for each individual or can you have a folder with both models photos in it but labeled correctly and have it still picked the right one after training on the folder? Another words, I have several people that I'd like to use often and I was wondering if I could just have a batch of photos of each one of them labeled correctly, then thrown into a folder zipped and then train that so I could simply call it by oh this should be a photo of Bill Jim and Joe etc. And thanks for the awesome tutorial.

[@techpiller2558]

Mona Lisa is so beautiful... Took my breath away.

[@riccardopala2502]

What if instead of being a common Instagram influencer, had she been an Onlyfanser? 😈🤭

[@yannmassard3970]

I love this mona lisa, totally understandable as in fact why he chose her. Fantastic work she is EXCEPTIONAL

[@jedeitor]

Very nice trick to use 2 character loras at the same time. I tried it and found that some facial features from one of them leak to the other and viceversa. It's a step forward but not quite there yet.

[@Otchengazoom]

I'm honestly speechless by the quality and how realistic the video is. A hundred thumbs up 👍!!!

[@cyberjungle]

💚

[@postandfly]

Thank you so much for your video! The only thing I can't seem to find is the link to access the part at 11:49 where we can mix public models

[@ainew002]

Thank you, have a nice day

[@fabiotgarcia2]

When it say "Combine this fine-tune with another LoRA". It´s actually mixing them. If you look closely, Leonardo and Mona Lisa are practically twin brothers. I tried several times to put two characters with completely different physiognomies but without success. :(

[@brexitgreens]

A tedious workaround: inpaint each face individually with the respective LoRA.

[@fabiotgarcia2]

@@brexitgreens could be the way.

[@OliNorwell]

yeah it's a fantastic tutorial, but yeah, the technique is 1000 times easier if your two subject happen to look like siblings!

[@DAOQuest]

May everyone thinks Mona is gorgeous, but I say Leonardo took my heart

[@OliverEnde]

WOW!! So great explained video!! This helps me a lot! I learned so much, thanks a lot!! I will try and use this new knowledge immediately :-) You got a new subscriber! Maybe I can mix that with my 3d animations.. Have a great day!!☺🤙

[@ainew002]

Thank you, have a nice day

[@MettameowChannel]

I love your channel! Thank you so much for the tutorial it's so useful and inspiring. Makes me wants to try it right away 🎉 Keep up the amazing work CyberJungle❤

[@cyberjungle]

💚

[@domedynamo]

These are going crazy time after time

[@studioterrible4444]

This is just great!

[@caldxm]

Wow that Mona Lisa part made me smile!

[@technobabble77]

Such a cool video. Awesome work!!

[@critiqsai]

Great tutorial and very well explained. We added it to our Best AI Tutorials.

[@cyberjungle]

💚

[@momirkoo]

Great work and advices! I use dalle3, sdxl, kling and suno ai but now i will reconsider some of my tools which i use to create ai music supported by cool videos

[@playthisnote]

Yeah nice trick. Those legends of Mona Lisa being Leo worked well for just adding a beard hehe

[@smithsmithington]

Jeez I want that short to be a miniseries. That would be such an awesome comedy twist to reality. Trippy and fun. Bravo.

[@ndcollins]

The reflection at 0:31 seems... a bit off. Where is the second person coming from, why is her reflection facing the opposite direction from her, and that second person has an arm coming out of her back

[@AndrySalim]

Super well done! This is amazing work..

[@iAmDonH]

Great video man! I’ve watched so many Lora training videos the last couple days and your video just opened my eyes. Thanks!

[@cyberjungle]

Happy to hear that! ✨

[@iAmDonH]

I like how you can add 2 characters using the trigger words in a prompt. Have you done research on splitting training into smaller sections. So you do a full face, then train again on the eyes, nose and mouth. Is that redundant 🤷🏻 and I was also wondering how it would handle tattoos if I train it on mine 🤷🏻

[@Notacet]

Wow cool, the tools are beginning to be there for short films etc. Are the consistent character tools abailable for pc/mac or are they online only?

[@JavierCamacho]

First time here... simply wao!!!

[@x4ms]

Wonderful. Yes, that's her.

[@leonardoa2479]

Thanks for the video, cheers from Brazil!.

[@cyberjungle]

Greetings to Brazil 💚

[@ezenwonsirim2620]

I would watch a vlog with Mona Lisa in it now, even as an AI her unique beauty is striking 😊

[@ahmedhelmy6125]

her smile 😍😍😍 0:37

[@iTrendMedia24]

Great tutorial! thank you !

[@cyberjungle]

💚

[@autoflujo]

Great tutorial! How did you create the voice over? Did you use ElevenLabs?

[@cyberjungle]

Yes I used Elevenlabs 👍🏼

[@nicdisan]

congratulations, beautiful tutorial and great experimentation, thanks

[@HaniTecBrix]

Excelent Content for Leo Mona Movie Clip :)

[@jgodvliet]

Awesome! Is having six fingers a trait of being a genius? :)

[@cyberjungle]

Yes indeed 😄

[@sandeepsaurabh8181]

After training Lora, Can we download models so that we can use it locally.

[@geromiuiboxz765]

1:22 5 finger + hidden thumb = 6 fingers = AI signature 😀 Saludos de 🇨🇱

[@bcraigcraig4796]

HI I love the short story You put together with all The visuals great Job and God Bless for all Your hard work Sir

[@alexbusoni]

Well, that's it, soon we will see films with the participation of Mona Lisa and Leonardo Davinci, with a new story, perhaps about the Titanic Part two: paintings come to life

[@davediamond]

Where else can we follow you. You've got the gold I've been looking for!

[@cyberjungle]

💚 I’m fully focusing on KZbin

[@vanillagorilla8696]

Could this be used to upscale old tv shows that are never going to get a remaster? I.e. The Outer limits from the 90s or DS9? I'd love to use this for projects like that.

[@whalewatchersa]

Beautifully done! Although Leonardo was left-handed.

[@bcraigcraig4796]

The workflow is really nice,I know I wish MJ could do this in one place maybe down the road they create some kind of studio place that we can use within MJ how much did all this cost?

[@fabiopobrazil]

Great tutorial! Thanks! I've been unable to get good results while combining two trained LORAs though. Facial features seem to always blend and/or MNALISA replaces the other LORA's features. Any way to prevent that (besides using the brackets)?

[@cyberjungle]

Hey Fabio, did you name your training images in the proper way? “Photo of MNALSA” “Photo of (your trigger word)” This is very important for good results 👍🏼

[@fabiopobrazil]

@@cyberjungle Hey! Thanks for your reply! I didn't actually. I had already trained a LORA for another project and the training images were simple named like "MNALSA" (without "PHOTO OF"), which worked well for that other project. I'll try retraining it and see what I get. Thanks again!

[@fabiotgarcia2]

@@fabiopobrazil Hey Xará, what you can see in the tutorial is that Leonardo looks very similar, they even look like brothers. The difference between them is their gender and one has a beard and glasses. In my attempts to make two completely different characters, one fat and one thin, the result was... twins.

[@fabiopobrazil]

@@fabiotgarcia2 Haha Exactly, Fabio. I also noticed that after reviewing the tutorial to make sure I had followed all steps correctly. There is even a setting in MNLISA page that allows us to set the weight of the additional LORA (the one I trained) when generating images, but sadly that didn't seem to affect results much. For now, I haven't found a solution.

[@edwassermann8368]

wow. incredible. thanks very much!

[@juxxcreative]

This is what I am looking for. Is there any possibility of using that locally on Comfyui?

[@mauriciovazquez7481]

great tutorial, but where or how I activate the use an extra Lora for a generation?

[@GodFamilyMoney]

Great Video. 💯

[@bcraigcraig4796]

When You show the characters faces grid do you need to crop each face out separately to train or can you use it as it is meaning the sheet of character faces on that grid

[@cyberjungle]

You need to crop each face out separately

[@bcraigcraig4796]

@@cyberjungleI am been creating a character using MJ and when I use full body shot I am getting whole body any ideas words

[@brexitgreens]

​@@bcraigcraig4796 Don't use full body images to train a face LoRA.

[@angryox3102]

When are we going to get locally ran open source video generation? I needs it in my life!

[@richardglady3009]

Lots of information. Thanks.

[@SebAnt]

Wonderful video, thank you so much

[@spadaacca]

I've used this LORA trainer, but it doesn't seem to give very consistent results :( I've used over 40 real clear face images from many different angles, but the results are lacking. Esp. compared to other LORA trainers that previously existed such as lucataco - but he doesn't have a Flux one anymore :(

[@Excapepath]

Can you make a video focused around a custom cartoon style character instead of realism please? For Flux Lora training on replicate. Thank you!! Love the videos I have learned so much from you.

[@cyberjungle]

Sure 👍🏼

[@Excapepath]

@@cyberjungle thank uou!

[@davediamond]

Yes! Would love this. Promised mum I'd help her with illustrations for her children's book.

[@nesrinunur2537]

Perfect video 👏👏👏👍

[@oligei666]

Nice videos! Support !

[@Luckx500S]

can you tell me if there's any Ai that can filter with clay?

[@Phonognomiks]

Good one! 👍🏻

[@RomanGen1]

Hello six fingers Mona, thankful AI still could be recognized in some cases

[@DenMarket-uc1nq]

That was great! I always wanted to see Mona in a modern context, including modern makeup. Leonardo, the creator of the painting, took a back seat to his own work - as so many of us do.

[@cyberjungle]

💚

[@a3103-j7g]

well then you could have always had a look at Denis Shiryaev's mona lisa renders four years ago.

[@vicentepallamare2608]

Would be helpful to provide links to models/tools which can be installed on PC - not paid models

[@bryant9967]

it seems to lack just a little bit from realism, the best achievement since sora, what we need in the future?: more natural staring, more capacity to mantain consistency in highly emotional scenarios, this is great news, soon we'll be able o create totally ia videos

[@dipennagda]

in training i am not at all getting the same input of image uploaded. I tried prompting playing with negative prompt etc but in vain. Anywhere you can help, i have followed steps as you have shown. Is it something to do with upscale.

[@mabelaye]

Very useful. Tks

[@cm1340]

UAU! Bel😍s!

[@Antonegoreviews]

How much per month in tools are we looking at??

[@intenSje]

altough the beginning scene with mona and leo is very beautiful and very impressive. i thought it was a bit cringe too: i think it would have been better if you had kept the timeframe of that scene in the era those people lived. otherwise: fantastic!

[@abdelhakyac7285]

To be honest, if you produced another channel with Mona Lisa and Leo and just imagined all the stories...... i would gladly follow .....

[@KblIndia]

Thank you for everything ❤️

[@cyberjungle]

💚

[@Jarbrain]

Thank you @cyberjungle for your video and it's pure gold information! I've set up two separate characters (Batman & Cat Woman, to help explain the issue) and tried to combine the two models into one. However, I have imported a second Lora (Cat Woman) into the other Lora model (Batman). The second Lora (Cat Woman) is influencing the other Lora (Batman) details to make a 50% blend between Cat Woman and Batman. How do I only import the second Lora without influencing the other Lora? Hope this makes sense! lol Thank you

[@brexitgreens]

A tedious workaround: inpaint each face individually using the respective LoRA. Very imperfect because the whole body and picture composition are affected in subtle ways - not just the face.

[@Jarbrain]

@@brexitgreens Yeah, I've noticed that the imported Lora affects all characters in view, not just the one. We can have two images using the same seed number... One without the Lora and another image with the Lora and then paint out areas that we don't want. It may not be ideal but it's something, I guess.

[@brexitgreens]

@@Jarbrain To be fair, it's not an effect specific to LoRA - all local image properties leak to the global image scope in generative AI to some extent. You can often observe it by generating images of Donald Trump and Joe Biden together. I guess this is due to the recursive (fractal) nature of image generation. I reckon it is one aspect of the "intelligence" of an AI model: how well it can resist this leaking.

[@brexitgreens]

@@Jarbrain That's clever too. Thanks.

[@chaseharley5853]

using the second "LORA"input makes the face mix with the first, meaning that the faces of both become 50% of the first character and 50% of the second. any ideas on how to do it?

[@cyberjungle]

Did you give some clues to the model in your prompt, clearly differentiating two characters?

[@chaseharley5853]

@@cyberjungle yes unfortunately that system doesn't work, I tried many times differentiating one character from the other but nothing, it takes a bit of one and a bit of the other in both characters. thanks anyway for the answer

[@ShaunJames-b9r]

This can be resolved with the prompts and adherence. What description are you giving? Put your image description up and I will separate the character s out for you

[@chaseharley5853]

@@ShaunJames-b9r I made many attempts and descriptions to make sure that the two characters were well distinguished, I also copied for example the prompt that he used in the video but nothing, one character always takes a bit of the traits of the other... I'm sure I did everything right with the prompt. does it work perfectly for you?

[@cyberjungle]

@@chaseharley5853 are captions of the images correctly written (process before training) ?

[@FranciscoFotoFilm]

Can I use a LoRA trained on Replicate with other accounts? In other words, can I train my LoRA with my face or a consistent character and share it for others to use with their accounts? If so, how does the process work?

[@cyberjungle]

You can deploy your trained LoRA on Replicate and make it public and share with others. You can also add it to Hugging Face and share with others again. Just ask in Perplexity and you will get many explainer resources 👍🏼

[@FranciscoFotoFilm]

@@cyberjungle tanks a lot!

[@webhack119]

Wow Image->Video

[@affaires-io]

Bravo !

[@muratsen-bq3nj]

2D veya 3D karakter üzerinde denedin mi hiç dostum ? Animatörüm ve çok merak ettim

[@cyberjungle]

Bununla ilgili çok talep geldi. Kesinlikle bir video yapacağım bu konuda ama teorik olarak aynı methodun çalışması lazım

[@muratsen-bq3nj]

@@cyberjungle Dostum denemek istiyorum fakat önce kredi kartını girmemi istiyor. Anladığım kadarıyla sonra fiyatlandırıyor. Ortalama 1 görsel için ne kadar ödedin .? Örnek vermem gerekirse 1 tane görsel yükleyip senin Davinci gibi 10 tane farklı pozisyonda görselini istesem ne kadar tutuyor ?

[@jeolman1]

my favorite part was the mirrors reflection of Mona applying make up at the ~:31 second mark

[@cyberjungle]

😄

[@PainlessTrader-h1u]

Your LDAVINCHI looks like the mans image on the shroud Tourin... IE Jesus - interesting

[@cyberjungle]

😂

[@brexitgreens]

I thought the same. But TBF we should have thought it decades ago rather than just now.

[@jawadkhelil5742]

Nice One, here we have an anagram for "Welcome to the jungle" it is a 1987 song by Rock band "Guns'N' Roses" from their album "apetite for destruction" The song's lyrics describes the gritty and often dangerous atmosphere of Los Angeles, particularly Hollywood, from the perspective of someone who is new to the city. Now "Jumanji : Welcome to the jungle" is a 2017 movie directed by Jake Kasdan and starring Dwayne the Rock Johnson (known for The Mummy Returns, the Scorpion King, the Fast and Furious movies, Hercules San Andreas, Baywatch, Rampage, Skyscraper Jungle Cruise ) and Jack Black (known for Shallow Hal, School of Rock, Ice Age, King Kong, The Holiday, Nacho Libre, Tenacious d in the pick of destiny, Tropic Thunder, Kung fu Panda) Now "Welcome to the jungle" in general is a slang term it means : -Encouraging someone whith a shared pity for example someone who is new to a very chaotic job and there are too many deadlines and the boss is always changing his mind the frustrated employee might say to the new person "Welcome to the jungle" Peace Next

[@gisfdlc9210]

Thank you so much🤝

[@cyberjungle]

💚

[@riccardopala2502]

He is more like Jesus Christ than Leonardo Da Vinci, but the final result is AMAZING!?! WOW

[@moneymanifestation9505]

All famous white guys look like Jesus😂

[@Dron008]

Very informative but it still requires a lot of steps and probably costs a lot. I'll wait for a moment when it is easier.

[@Paroissien]

Such talent! Thank you for the very first in flesh look at Mona in history since 1510+ !

[@rajsingh-bg3jr]

nice man

[@marciaribeiro3128]

Ela é idêntica ao quadro da Mona Lisa

[@msza5560]

How to search for a model you created in replicate ?

[@cyberjungle]

Click on your name and go to homepage. Trainings = Models you created Predictions = List of Images your created

[@AIChandu77]

i dont find extra lora option

[@cyberjungle]

First you need to train a LORA and after that you can add second one. Or alternatively you can check reflux.replicate.dev

[@CarlosAlvarez-cj8zg]

Would this work for cartoon characters?

[@cyberjungle]

Yes 👍🏼

[@arthurkawada9610]

amazing

[@thomasproshowski7538]

Dear Admin , can you please make the people on the video talk !! Than you

[@ChcxCivd]

can consistent character and face made with runway/hailuoai?

[@cyberjungle]

There is Custom Model option on Runway. academy.runwayml.com/ai-training/using-your-model There you can train a character’s face or even a product/item but it’s as good as Flux Lora training (for now) Hailuoai doesn’t have this option yet.

[@residualroyaltyacademy]

what is the real flux AI website? there seems to be 2 diff ones

[@cyberjungle]

blackforestlabs.ai

[@GrooveStation-n6c]

Mona Lisa really was hot I guess, love it

[@AllenReadsLit]

What about consistent lip syncing?

[@cyberjungle]

Not possible with Kling. It’s possible with Runway Gen3 👍🏼

[@AllenReadsLit]

@@cyberjungle I look forward to seeing you make a video about this, perhaps in the near future!

[@OceanWalk7]

I'm still torn about doing anything with my own face. I don't want my face to end up on other people's videos. While I think this is so cool, I am scared of a world where anybody can make realistic videos of anybody. Most people don't even have the AI literacy for text and photos yet. And it's not like the world is just a happy paradise. A lot could go very wrong with people using this stuff in a bad way.

[@cyberjungle]

100% agreed 👍🏻

[@msza5560]

How to find a template I created

[@yusefbn9522]

The bad thing about Kling is you need to wait like 24h to get your results

[@cyberjungle]

It’s definitely not cool :(

[@666regicide]

@@cyberjungle It's like shooting on film in the old days 😸