MASTER LORA OBJECT Training FOR SDXL! QUICK & EASY!

Подписаться 150 тыс.

Просмотров 30 тыс.

50% 1

In this video, I'll walk you through the process of training OBJECTS with LORA FOR SDXL! Just like before, I've poured hundreds of hours into research and dropped hundreds of dollars on cloud computing to bring you the go-to guide on LORA object training, tailored for both newbies and pros. SDXL makes object training a breeze if you're armed with the right know-how and settings. By the end of this video, you'll be a wizard at training any OBJECT using LORA FOR SDXL with the Kohya GUI tool.
What do you think of the LORA Object Training? Let me know in the comments!
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
SOCIAL MEDIA LINKS!
✨ Support my work on Patreon: / aitrepreneur
⚔️ Join the Discord server: bit.ly/aitdiscord
🧠 My Second Channel THE MAKER LAIR: bit.ly/themakerlair
📧 Business Contact: theaitrepreneur@gmail.com
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
✨ PATREON LINK: / aitrepreneur
Kohya ss GUI: github.com/bmaltais/kohya_ss
SDXL: huggingface.co/stabilityai/st...
1-Click Stable Diffusion A1111 installer: / 1-click-install-86931758
1-Click Kohya_SS installer: / 1-click-install-87532002
Training Object presets: / object-lora-91936208
Exclusive LORA models: / 5-new-lora-91936651
Background remover: www.photoroom.com/tools/backg...
LORA Playlist: • LORA SDXL
General LORA training video: • ULTIMATE SDXL LORA Tra...
Previous Style LORA training video: • SDXL LORA STYLE Traini...
Clothings LORA Training video: • TRAIN Perfect SDXL LoR...
scale_parameter=False relative_step=False warmup_init=False
Runpod: bit.ly/runpodAi
Runpod Template: runpod.io/gsc?template=ya6013...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
►► My PC & Favorite Gear:
i9-12900K: amzn.to/3L03tLG
RTX 3090 Gigabyte Vision OC : amzn.to/40ANaue
SAMSUNG 980 PRO SSD 2TB PCIe NVMe: amzn.to/3oBR0WO
Kingston FURY Beast 64GB 3200MHz DDR4 : amzn.to/3osdZ6z
iCUE 4000X - White: amzn.to/40y9BAk
ASRock Z690 DDR4 : amzn.to/3Amcxph
Corsair RM850 - White : amzn.to/3NbXlm2
Corsair iCUE SP120 : amzn.to/43WR9nW
Noctua NH-D15 chromax.Black : amzn.to/3H7qQSa
EDUP PCIe WiFi 6E Card Bluetooth : amzn.to/40t5Lsk
Recording Gear:
Rode PodMic : amzn.to/43ZvYlm
Rode AI-1 USB Audio Interface : amzn.to/3N6ybFk
Rode WS2 Microphone Pop Filter : amzn.to/3oIo9Qw
Elgato Wave Mic Arm : amzn.to/3LosH7D
Stagg XLR Cable - Black - 6M : amzn.to/3L5Fuue
FetHead Microphone Preamp : amzn.to/41TWQ4o
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Special thanks to Royal Emperor:
- Totoro
- TNSEE
- RG
- Judy Godvliet
- Gluthoric
Thank you so much for your support on Patreon! You are truly a glory to behold! Your generosity is immense, and it means the world to me. Thank you for helping me keep the lights on and the content flowing. Thank you very much!
#stablediffusion #sdxl #lora #aitraining #texttoimage #imagegeneration
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
WATCH MY MOST POPULAR VIDEOS:
RECOMMENDED WATCHING - All LLM & ChatGPT Video:
►► • CHATGPT
RECOMMENDED WATCHING - My "Tutorial" Playlist:
►► bit.ly/TuTPlaylist
Disclosure: Bear in mind that some of the links in this post are affiliate links and if you go through them to make a purchase I will earn a commission. Keep in mind that I link these companies and their products because of their quality and not because of the commission I receive from your purchases. The decision is yours, and whether or not you decide to buy something is completely up to you.
Timestamps:
00:00:00 Train SDXL with LORA
00:00:28 Koya SS GUI
00:01:02 Image Dataset preparation
00:01:45 Remove the background of an image with Photoroom
00:02:36 Upscale images sharper with Stable Diffusion
00:04:19 SDXL LORA Training Folder Creation
00:04:36 Instance Prompt & Class Prompt
00:05:19 LORA Repeats Training Steps
00:05:50 Folders Tab
00:06:06 Image captioning for LORA training
00:07:32 The right SDXL LORA Training Parameters
00:07:47 Train Batch size
00:08:13 Number of Epoch Training
00:08:30 BF16 & FP16 & Cache & LR Scheduler & Optimizer
00:08:57 Learning Rate for SDXL LORA
00:09:17 SDXL Image Resolution & Buckets
00:09:39 LORA Network Rank & Network alpha
00:10:06 Color augmentation & Flip Augmentation
00:10:14 LORA Training
00:10:27 GPU Solutions & RunPod
00:10:39 Select the right LORA Model
00:11:24 How to generate images of a person using the object?
00:12:26 Conclusion on object LORA Training

Опубликовано:

6 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 61

@Aitrepreneur 8 месяцев назад

HELLO HUMANS! Thank you for watching & do NOT forget to LIKE and SUBSCRIBE For More Ai Updates. Thx

@beansstark1389 8 месяцев назад

Kindly share a razorpay link , Patron is banned in our country India

@danielkaiser4412 3 месяца назад

Can you please make a video showing where hands are placed. hands in the hair, hands on the stomach or on the hips or combinations of these? I've been trying out Lora's for two months now and can't get it right. It seems to be quite difficult and complicated. It would be great if you could help me. Sorry, I don't know what to do... Thank you

@carolineito9312 8 месяцев назад

Yaaay you are a lifesaver! Thanks for putting so much effort and study to be here sharing knowledge with us ❤

@Aitrepreneur 8 месяцев назад

Happy to help ;)

@pointlesspos8440 7 месяцев назад

What i needed for product placement in e-commerce, perfect.

@aimademerich 8 месяцев назад

Finally!! Objects, thank you!!

@tomkramer05 3 месяца назад

Excellent!, thank you for the insights

@kernsanders3973 8 месяцев назад

These are awesome! Would love to see one on concept training. Like getting a character to do something a bit more complex than can be done with just controlnet. That and clothes I use to struggle with but your clothes training tutorial helped a ton.

@madrooky1398 7 месяцев назад

I am not saying it impossible with a Lora, but depending what you are aiming for it may contains so much context that it is better to apply directly to the checkpoint. Contextualisation of concepts is they key here, and that requires a well crafted data set and lots of patience to train with it.

@kernsanders3973 7 месяцев назад

@@madrooky1398 Think it probably really depends on each concept. Because theres tons of concepts on civit. Some of them are mind boggling good. Would love to see the dataset that were used to train them. Some of them barely interferes with the character and style, just mostly applies the concept. How they managed that? I have no idea. But of course some concepts are extremely complicated to train that it is better to train it in a checkpoint than a lora.

@baziztsahy3927 8 месяцев назад

Great video!

@Aitrepreneur 8 месяцев назад

Glad you enjoyed it :)

@andrewcarkhuff2640 5 месяцев назад

Wow thanks! I will use this to help me put good swords into my Ai animations.

@SinfoniaTrueClanMex 8 месяцев назад

Next video how to make animations with stable diffusion, without too much flickering, please!

@Aitrepreneur 8 месяцев назад

Sorry I can't make magic, the tech is still not there yet

@mentocthementalist5917 8 месяцев назад

Thanks again Aitrepreneur !! please make a video on training a location; if that sounds like a good idea to you. This is something I was thinking about, like inside a specif building as an example

@mehmetturk8471 8 месяцев назад

I used your settings, everything is fine, but the labels on the products are distorted, which settings should we upgrade to clarify them

@KriGeta 7 месяцев назад

how can we make anime comic like dragon ball using this?

@vendacious 8 месяцев назад

OH MIGHTY FRIEND COMPUTER! Ever since your first Kohya-SS tutorial, I have trained many LoRAs and I am grateful! *gets on hands and knees* But please consider finding a way and teaching us how to create a our own ControlNet LoRA models with image pairs, as the current workflow and hardware requirements for creating custom ControlNets is too complex for my tiny human brain and pathetic consumer-grade hardware. -Your Loyal Citizen

@scem-currentspecialist 7 месяцев назад

Hi.... I need a big advice.... I am using comfyui in my laptop, I have a perticular product I want in my image generation, that is my Electric vehicle... I can take multiple images of the product... But is it worth it? Can I use it for posters, banners, pamphlet etc? I am an electric vehicle dealer in my small city, I am trying to get AI help in marketing my product. For example I want a poster where santa is riding my scooter for my Christmas poster. Is it possible to make that happen? This would be a great help. Thank you.

@balanceyourlife5973 5 месяцев назад

You're just aweasome! I am wondering if there is a way to add more Images to a trained lora, do you have any clue if this can work or do i have to retrain it again and again and again?

@Crisisdarkness 6 месяцев назад

Please, I hope you or someone can help me. I wanted to make a Lora, for the new SDXL turbo model, do you know the resolution required for the input images? Is it 512x512 or 1024x1024? And what are the key parameters? Could you make a tutorial? Thanks.

@BM-pt6sy 8 месяцев назад

Can you make a tutorial on converting objects to mechanical design versions? Like turning an arm into a cybernetic version with a unique design.

@jurandfantom 8 месяцев назад

Anyone else struggle with item floating in air out not blending with scene? Looks like in my case (where all source pictures of object have white background) i got such issue.

@emskidon3084 8 месяцев назад

Guys, i need help. What is the best option ro train SD for more realistic Instagram posts, like backgrounds and scènes. Restaurants, hotels or in streets. I want to use this in combination with an character/person LoRa.

@researchandbuild1751 4 месяца назад

I've been trying to train electronic components, but it's not going too well yet. I think my image resolution was too low and that could be why

@hagarborges9382 8 месяцев назад

As a suggestion, how about Lora (or any other solution you might have) for fighting scenes like martial arts or even wrestling? I find it very difficult, especially for the person at the receiving end, even using controlnet.

@Aitrepreneur 8 месяцев назад

Since SD still have trouble generating correct body positioning, the only way to generate those images is with controlnet, no lora or DB is gonna change that yet, we might need to wait for the future SD models

@vilainsinge5282 5 месяцев назад

I want to train a LORA based on a face mask that i've crafted. I dont know if i should train it a a person - with me wearing it, or as an object - to use it on different people. If i train it as an object, i considered put the reference images of me wearing it and specify that it's worn by someone in the description. Any advices ?

@kimjongoof5000 5 месяцев назад

How different is this tutorial process from training on a cartoon character? I want to make Advance Wars style sprites of "Geumsaegi" from Squirrel And Hedgehog

@vi6ddarkking 8 месяцев назад

An interesting question. Could this be trained to Add tails or animal years to a character.

@Aitrepreneur 8 месяцев назад

this is more like a concept or cloth training instead, just get a bunch of images of that and train it like a cloth/outfit

@stas_lu 8 месяцев назад

write who has what training speed on the local computer? I have a 3060 12 GB graphics card

@tetsuooshima832 8 месяцев назад

Guys, I'm going to be slightly off-topic but does anyone know a tool like BLIP/CLIP Captioning in batch like in Kohya_ss but able to put the tags in the pictures instead of an external text file? I'm looking for this in order to organize my own pics, would be nice not having to do that manually

@sharkeys 8 месяцев назад

I haven't got time to test it but Chat GPT can easily help make this possible. I've used to for creating many dataset management programs before. try the prompt "could you help me make a program that puts tags into png metadata from a text file of the same name"

@fpvx3922 7 месяцев назад

Sadly my Computer can't train SDXL... wish you showed how to do Lycoris, Loconn, Lora on SD 1.5 I have a dog that I want to have as a model and all guides only have explanations that do not fit for the pet.

@tds_FrankOfficiel 3 месяца назад

i use Stabilty Matrix and it doesn't work, and I have absolutely no idea why 😭😭😭

@oliverp764 8 месяцев назад

Can you please make a guide for clothing training? This would be cool

@Aitrepreneur 8 месяцев назад

already did ;) Check the previous video :)

@Ricrosmir 8 месяцев назад

This works only for sdxl right?

@Aitrepreneur 8 месяцев назад

No this works for SD 1.5 also, the principle is the same :)

@flareonspotify 8 месяцев назад

You should make make a simple version of your SDXL train Lora video there’s to much information to watch a very long tutorial

@banzai316 8 месяцев назад

Very useful, AItrepreneur!

@Aitrepreneur 8 месяцев назад

Glad it was helpful! :)

@tetsuooshima832 8 месяцев назад

"Now let's use the upscaled images" * picks the BEFORE folder * - wut?

@Aitrepreneur 8 месяцев назад

Woops :D

@tetsuooshima832 8 месяцев назад

@@Aitrepreneur you also said "786,786" for max resolution while I would expect "768,768", but I didn't want to be like a bully pointing it out lol

@Laszer271 8 месяцев назад

Artists are sad because AI can steal their jobs. Meanwhile, as an AI researcher, I'm a little sad that everything has become so easy. I remember testing various GAN architectures as well as designing our own with my friends in 2018-2021. Nowadays everything is so easy it's not even fun. I have to pivot to a different field as image generation isn't interesting for me anymore. I wonder if there are people that feel the same.

@banzai316 8 месяцев назад

Switch to ComfyUI for a little challenge

@tetsuooshima832 8 месяцев назад

Nope. It's like wanting to do rigging and retopology again. Ah wait, we still have to do those

@randomguyfrominternet 8 месяцев назад

I think that artists are gradually starting to calm down. Image AI hype is fading as people are finding that: 1) It is not possible to control it as much as they would like because then the results are not so good, or it takes so much work that it discourages most users. AI image generation works the best when it controls you, not the other way around. 2) It won't make you a real artist like every user thought. You may collect some numbers on IG, but otherwise no one from companies or the professional market cares. 3) 95% of all uses and creations are incredibly repetitive NSFW, weird fetishes, or generally just posing women who don't appeal even to the adult sites anymore. 4) The internet is flooded with the AI stuff so much that the people actually start to hate it. 5) After some recent huge updates like SDXL, people start to realise that the progress speed of these technologies won't be that extremely fast as before. Even more so as other researchers and creators are finding ways how to protect the art. I am actually designer but I went with the hype and spent several months playing with SD1.5. But man, everything is so similar after a while. Especially if you compare with others. It's like the whole internet content is created by 3 people now. But I am grateful that I did. I bought a drawing tablet after that and started to learn how to properly draw. So the AI actually helped me to become better artist after all :)

@banzai316 8 месяцев назад

@@randomguyfrominternet There’s so much that goes into lighting, 3D perspective and just creativity. AI can never replaced a real artist. It helps me get started, now I realized lighting is a work of arts

@tetsuooshima832 8 месяцев назад

@@randomguyfrominternetexcuse me, people especially artists making stuff for a living didn't wait to get flooded to start hating AI. They hate it with a passion from the start xDD

@user-im1iy4ir8i 4 месяца назад

probably the worst advice for training a proper lora. congrats

@user-id7pe9wo6i 3 месяца назад

do you have a discord

@sitr2516 8 месяцев назад

so i guess there isnt much happening in the world of AI? these last few videos have been kinda... well... not much new news i guess. whats going on out there?