FLUX + LORA Full Course (Uncensored Super Photorealistic Images With Your Own Face)

[@cyberjungle]

Midjourney V6.1 - Photorealistic Cinematic AI Photography Style Guide: cihanunur.gumroad.com/l/midjourneyV6-style-guide

[@Hadia-o4h]

can u please let me know how can we train the model to add some traits to the faces like scars with stitches on the cheeks because the current scars created are not exactly as i want for example it shows a small red line without any stitches so please help

[@JérômeTienne]

🎯 Key points for quick navigation: 00:00 *📸 Introduction to Flux Laura* - Overview of Flux Laura's capabilities in AI photography. - Explanation of four main sections covered in the video. 00:14 *🎨 Using Flux Laura on Replicate* - Demonstration of image creation on the Replicate platform using specific prompts. - Comparison of different prompts to achieve varied realistic images. 02:11 *💡 Exploring Paid Options for Flux Laura* - Discussion of additional platforms like Free Pick Picasso and File.a for using Flux Laura. - Explanation of subscription options and limitations for generating images. 04:40 *📚 Training Custom Models on Replicate* - Step-by-step guide on training personal models using your own images. - Importance of image quality and labeling when creating custom models. 08:29 *🧑‍🎨 Running Your Trained Models* - Demonstration of running a trained model and evaluating its output. - Analysis of how changes in guidance settings affect the realism of results. 11:11 *🐾 Training Models on F.A* - Overview of training a model using another platform, F.A, focusing on a pet image. - Recap of the credit system and costs associated with the platform. 14:05 *🖼️ Image to Image Prompting* - Explanation of using image-to-image prompting features on different platforms. - Discussion on achieving desired details and expressions in generated images. 16:01 *🦁 Inpainting Techniques* - Introduction to inpainting using Flux models for creative extensions. - Addressing challenges faced while trying to achieve specific features in images. 17:04 *🎥 Creating Videos from Images* - Guidance on transforming Flux images into videos using Clink and Runway platforms. - Emphasis on proper prompting for effective video generation. Made with HARPA AI

[@marian_c90]

As I understand, on image to image you used an 0.8 prompt strength, almost ignoring the image prompt. That's why you got that result. Besides that, image to image function doesn't work like that. You need to use the lora of your pet to get proper results of him, and this function only will trace the input image using it as a guideline (of shapes and colors), but not using it as a source of the character

[@marian_c90]

on inpainting I suggest you to edit the input image first, drawing a basic shape of what you wanna get, and test different strength values. In this case, I would paint a black shadow with lion shape, and the prompt would be something like 'lion shape backwards, walking by a street with sunset bla bla bla..' Sorry for my english

[@HettieDeleon]

kahma AI fixes this (AI Headshots). Uncensored super photorealistic images.

[@SecondLifeAround]

Midjourney's Reaction: "What the FLUX!"

[@Avembe]

no u didn't....

[@JohnVanderbeck]

I still think MJ images are unique and in most cases better still. There is a lot of "magic" that they are doing to the prompts that is hard to replicate unless you are VERY good with prompting.

[@Avembe]

@@JohnVanderbeck i had this impression and now i just read what u saying ,its pretty obvious that their own prompts must be from another level were probably not aware of or unable to imitate

[@JohnVanderbeck]

@@Avembe yeah basically they take what you give them, then apply some ML magic on top of it to create a new prompt that is then fed into their model. So the idea of a "Midjourney killer", which don't get me wrong I'd LOVE to have one I could run offline on my own machines, is really about MORE than just having a really good model.

[@humanoide_existenz]

The problem with your trained Flux LoRa model was probably that your trigger word was far too general. If you had used something like "ohwxman", the trigger word would not intersect with other data sets within the flux model.

[@cyberjungle]

I think trigger word cyberjungle was quite specific and not so generic no? But I will consider your suggestion next time 👍🏼

[@elizagarcia8799]

Is there a Interface or something for Training FLUX LORAs in Local PC?

[@a.akacic]

you are flat brained. you clearly have no idea what you are doing. civitai users are cringe

[@kilo_368]

love you brother you are top tier when it comes to this digital art, Iam so thankful somehow I found your channel, you are the only one when I watch KZbin videos I feel like I have learnt something phenomenal.

[@VaibhavShewale]

i train the model with my face, for first image it generated perfectly but from 2nd image it went downhill and from 5th pic it started generating me as a women

[@cyberjungle]

Yes I had similar experience. Results improved when I specified my gender in the prompt

[@ryudious]

Must be Canadian 😂

[@J3R3MI6]

@@ryudiousperfect 😅

[@mehditayshun5595]

Canadian ?

[@me-cm8or]

It showed you all the possibilities

[@blacksheep480]

Ideogram generates far more effectively than Flux... I have checked one prompt for it: "A vintage café with warm, soft lighting, wooden furniture, and a comfortable atmosphere. Large windows display raindrops streaming down the glass, with a wet street outside. Inside, customers are sipping coffee, reading books, and engaging in hushed conversations. The aroma of fresh coffee and baked goods permeates the air." For this particular prompt, Ideogram seems to provide a more vivid and detailed description.

[@cyberjungle]

Which guidance value did you use for FLUX? I observed in my tests that guidance makes a big difference for flux

[@blacksheep480]

@@cyberjungle Thank you for your response. I appreciate your time and consideration.

[@blacksheep480]

​@@cyberjungle Could you please provide more details about the different AI image generation tools you are interested in? I'd be happy to share my thoughts.

[@patrickzupanc1795]

Thank you for the great video!

[@cyberjungle]

💚

[@sumsumab1809]

Thumbnail is every just chatting stream 😂😂😂

[@ahaa1233]

When inpainting you need to describe the existing image in detail as if you were recreating it + describe what you want to add

[@BuddyViewer]

Kling Ai looks crazy!

[@DaphneMarina88]

Kling is pretty fun!

[@Anya.petrova.official]

Great value brother thank you!

[@Huang-uj9rt]

Yes, I think what you said is great. I am using mimicpc which can also achieve such effect. You can try it for free. In comparison, I think the use process of mimicpc is more streamlined and friendly.

[@arieltrivino1732]

can i use this for commercial use? Let's say I want to charge my friends $20 to create Instagram images of them in different places or do I need a specific license?

[@cyberjungle]

hey as far as I know it's personal projects only. For commercial usage you probably need a special commercial license. You can contact [e-mail] for 100% clarity 👍🏻

[@koufdell]

dat roboflow supervision guy !!!

[@muharremacar5255]

When I click on create training it says "You need to have a payment method set up to run this model."

[@pranithkumar8924]

for training it takes aroun $5, after that the generations are very very very inexpensive. like 20 cents per image or something like that

[@muharremacar5255]

@@pranithkumar8924 After creating Lora, we can download it and use it locally or elsewhere, right?

[@sirishkumar-m5z]

With SmythOS, developers can leverage advanced AI tools and libraries that are optimized for the OS, leading to faster experimentation and iteration. This streamlined approach accelerates innovation in the AI field.

[@GenZero-x6m]

This is so amazing, thanks for sharing.

[@pigreatlor]

man thanks i love you for this

[@KAVaviation]

what is the best hyper/lcm model to use?

[@onrsaygi]

eyvallah abi işime yaradı

[@cyberjungle]

💚

[@AsokaTw-mz3lr]

how we use the same faces throughout different images? the main problem of AI is each time the image is generated, the person is a different one.

[@Avembe]

Can you specify that you want the very same face on each new generated image ?

[@bratan007]

I'm running Flux on my local computer. Is there a way to get it working with LORA locally?

[@MiracleMan-ol6gd]

I'm far from an expert, but since Comfyui can handle both LoRA and Flux, you should be able to combine these. I'm still a novice, using only Fooocus and Comfyui currently. Because I'm impatient and mainly copy workflows from others.

[@pshoweu]

Great walkthrough, thanks. 👏🏻

[@yeoworld]

Thanks. That was interesting.

[@jensk9564]

way more images of your dog than of yourself -> means: you heart is on the right place

[@cyberjungle]

💚💚💚

[@noelomondi4849]

Do you know whether it’s possible to get image to image on Replicate?

[@ChyaBra]

Thanks for the video. It works great locally in Comfy. I can't seem to get an illustration style though, it just does photo realistic even with anime loras and models. Any advice on getting a more stylized cartoony output?

[@cyberjungle]

I would try Flux Pro (or Flux Dev) without Lora for illustrative art style

[@gamingdayshuriken4192]

i wanna lose myself in your jungle dude ! thx great work here !

[@TheBlueRage]

Nice video. I just started using ai tubo and is interesting software.

[@kevincoyle3388]

Thanks for the video - it was really helpful. Is there any way you could explain to me how the image-to-image works on fal? Do I need to manually add the flux-realism lora and, if so, where do I find the path/URL?

[@juicy3397]

Hi thanks for you video , I want to ask you a specific question if u can answer me , did u think we can get really good "2D pixel art sprite sheets character animations" ? Did you know a good prompte to have a real good stuff ?

[@SecondLifeAround]

Cyberjungle-Core!

[@ak_mits]

Çok teşekkürler! Tercihiniz Replica mı? Local PC de training ve kullanim yapabilir miyim offline olarak?

[@cyberjungle]

Replicadan memnunum evet. Lokalde de kurulabilir ama ortalama üstü bir PC gerekiyor👍🏼

[@bcraigcraig4796]

By creating your own model does the company now own all your images that you will use to train?

[@kelvynbeats4371]

Good question

[@pennywise5095]

Well you use all other images without any reservations

[@johnmastodon-z2f]

Insane how much this thumbnail looks like olivia tied..

[@bcraigcraig4796]

what is your formula for creating great prompts Flux Lora ? Thanks

[@cyberjungle]

Flux Lora excels in natural language understanding. So you can use this formula: [PHOTOGRAPHY TYPE] [SHOT TYPE] of [SUBJECT/ACTION], [LOCATION] [YEAR] [FILM STOCK] [LIGHTING] [FASHION] [EMOTION] [COLOR] Not every element is essential here but for me photography type, shot type, subject/action and location make huge impact for photorealism

[@bcraigcraig4796]

@@cyberjungle does this Prompt Formula work with MJ images?

[@24-7gpts]

@@bcraigcraig4796 yeah

[@bcraigcraig4796]

@@cyberjungle how this A portrait,close up,woman,looking at the camera,park,Golden hour,expression fear I cannot seem to get her to show fear

[@StephenRayner]

With the hugging face in painting it can only paint over something similar that was already there

[@obiwank2]

Thanks for this video. I can't upload the zip file. Does it need to have a specific name or characteristic ? Thanks

[@cyberjungle]

nope not really. What kind of error do you get?

[@obiwank2]

@@cyberjungle Actually, nothing happens. It seems to take the file I drop into the window, but nothing happens. I refreshed the page, but it's the same.

[@franciscobutte]

Can you download the file of your trained model so you can use it locally?

[@roydadon9889]

Why when i try to train my model it says my learning rate value isn’t valid even tho i use the default?

[@Techonsapevole]

great is possible to use the custom LORA also with Flux Schnell ?

[@forceawakened8084]

yes

[@ortizgab]

Hi! Thanks for the info. Can I download the trained model from any of these sites?

[@cyberjungle]

You can download models and run locally from civitai.com

[@ortizgab]

@@cyberjungle Yes, but can I download the models trained with my images?

[@Monotv67]

perfect 💝💝💝💝💝💝

[@k2an]

from TR right:) nice work btw liked and followed

[@zhonkvision]

Thank you

[@MaxiT0070]

Nice. How to incorporate real product in an image ?

[@mr.entezaee]

Why is my zip file not uploaded?

[@Goepf]

When i try to upload my .zip file it does nothing. Does anyone Know how to fix it?

[@cyberjungle]

Which error do you get? Did you press Run to start training?

[@mr.entezaee]

@@cyberjungle Same for me. It does not allow the zip file to be uploaded

[@ErickGanem]

Same for me here. Did you came up with a solution?

[@ErickGanem]

@@cyberjungle There's no error when drag the zip or pick it from the popup windows it doesn't respond or send any message

[@oberon2230]

Does anyone know if there is an AI that takes your face and creates an environment or theme around you without making your face look different? So if I want to say, take a photo of my son, make sure his face is not changed, but his body and environment would be : Jungle setting, military outfit with a machine gun, hyper photo realistic. There are so many AI tools out there I can't figure out who does it this way. Everyone messes up the face, or makes your face look like a CGI look

[@cyberjungle]

Take a look at the video here. I think it will be helpful for your case: kzbin.info/www/bejne/o4a1aXaLoqmVpZI

[@oberon2230]

@@cyberjungle Thank you so much, outstanding videos you have! Consider me Subscribed :D

[@burskyozbekov]

Can you explain how to use this with Forge? Tesekkurler

[@cyberjungle]

Okay noted for the next video 👍🏼 Rica ederim

[@Shwemy99]

Replicate is paid now?

[@fernandomartin8147]

Hi there @cyberjungle liked your channel. Im wondering if you can merge a ckeckpoint you trained with a flux checkpoint (like you can do with Forge UI).. is that possible? Thanks

[@raztube90]

What is the use of : in runway ?

[@cyberjungle]

It’s just a simple seperator between camera movement and event happening in the video. Example: Over the shoulder shot: the man walks in the forest

[@raztube90]

@@cyberjungle so no difference between : and ,

[@mjklein]

Anyone remember SIM ONE? Yeah, we have arrived.

[@giuseppedaizzole7025]

where is the uncensored thing?

[@cyberjungle]

You can prompt Flux Lora uncensored without any errors or limitations. I can’t demonstrate that on KZbin

[@micbab-vg2mu]

thanks:)

[@sagarkapasi099]

Cool

[@pointy7771]

yahu aksandan dolayı aha Türk, bizden diyip, dönüp ana sayfaya ve fotoya baktım... I waz rayt. :) eline sağlık. Kohya gibi bir trainer çıkmadı mı hala? 10 dolar fln bizim ülkede çok biliyorsun :)

[@HemantkumarSharma-n1s]

which coupon code were you talking about?

[@cyberjungle]

Coupon Link: replicate.com/invites/572ae3c6-1385-41b3-afdc-1e99e11836b8 Found in Matt Wolfe's video here: kzbin.info/www/bejne/laPNpaJqnt1mqKM

[@HemantkumarSharma-n1s]

@@cyberjungle Thank you soo much. Keep up the great work. Also, can we use this trained lora in COMFY UI?

[@HemantkumarSharma-n1s]

@@cyberjungle Hey could you send a new link, the above seems to be expired

[@chrissoba5104]

Is free ?

[@juanracordoba6943]

works on Mac??

[@cyberjungle]

I have only used online trainers: fal, replicate etc

[@shamo84]

@19:11 incredible

[@AkshayKumar-vd5wn]

Can you use this in Google Collab?

[@cyberjungle]

Yes 👍

[@ypsilantiavanti5603]

Even more realistic if not all women have perfect teeth. Who says that a huge hole in the teeth is not normal? 🤣

[@nesrinunur2537]

❤

[@pelufaz8435]

damn I need to pay $10 dolla

[@FranckSitbon]

Camera is not static. Gen3 is the best animation tool, but he has a lack of comprehension of camera.

[@pastuh]

Teeth made photo realistic 😅

[@liongbigfoot7939]

It's not free right?

[@cyberjungle]

Free on Hugging Face 🤗

[@georgepenzenik9994]

But the images still do not pass the test and are still easily detectable as AI Generated.

[@7sonderling]

wtf... Flux inpainting link on huggingface is just bs... doing nothing except examples.

[@ilgiligili]

kardesime bir layk atayim

[@mlnima]

lol, dude learn some python and buy a 24gb vram GPU, I am creating lora models on my pc and use it unlimited with freedom XD

[@careerdog]

Can u do it on a 32gb laptop with 8gb vram on a mobile 4070 gpu?

[@mlnima]

@@careerdog no, you need high VRAM to train a lora but it hell worth it. try to find some 3090 which is not expensive atm

[@pigreatlor]

is it possible to use flux pro as well on your pc? and do you have a tutorial video for this anywhere? ive already installed comfyui + flux dev and it worked good so far locally

[@mlnima]

@@pigreatlor pro is not available for download but you can use via API, red the docs of the model

[@ishaat_plays]

How about 4080?

[@red_head8662]

Brother, just think about it, AI can create realistic games frame by frame. 🤣

[@alp2387]

Are you Turkish bro? Your accent looking like so Turk..

[@ermanmanyasli]

türk ingilizcesi nerede duysam tanırım :)

[@cyberjungle]

😄

[@johnt697]

What accent is this?

[@noobdestroyer5454]

Oh no

[@CinematicAIPlayground]

THX Mate - this helped me to create an AI video (kzbin.info/www/bejne/nKKnZaCqlLWHnKMsi=xdm3boDQSFdsS0l4) with my boyfriend as a superhero within 6h. Amazing possibilities.

[@cyberjungle]

Hey this looks fantastic! ✨ Great job 👏🏽

[@noobicorn_gamer]

1:17 The image looks like taken with a real camera My man needs to look at more real flickr photos of those era...

[@bilincinontolojikizdirabi]

aksanından türk olduğunu anladım hocam tebrik ederim, ben de bu alanda kanal açmayı düşünüyorum. instagramdan takibe başlıyorum

[@Jonki125]

good video, thank you!

[@cyberjungle]

💚