LoRA Training in UNDER 5 MINUTES in Stable Diffusion! THIS IS CRAZY!! (Create Comic Book Characters)
ฝัง
- เผยแพร่เมื่อ 1 มิ.ย. 2023
- In this episode, I will explain LoRA training in UNDER 5 MINUTES in Stable Diffusion! THIS IS CRAZY!! (create comic book characters).
Do you want to create comic book characters using Stable Diffusion LoRA? In this video, I'll show you how to create comic book characters using stable diffusion LoRA in less than 5 minutes!
This technique is simple and easy to use, and it can be used to create scenes and characters for your comics.
It's made on the Wacom Cintiq Pro 32, Macbook Pro M1 Max, Dreamlook.ai, RunDiffusion, Atem Mini Pro, Adobe Premiere Pro, After Effects, and Adobe Photoshop.
-------------------------------
SUBSCRIBE
▶ handle: / @levendestreg
▶ You can subscribe to our channel here: th-cam.com/users/levendes...
▶ Promo code for Dreamlook.ai - use "LEVENDESTREG" for 30% discount for up to 100 redemptions, without a time limit.
▶ RunDiffusion Promo Code: levendestreg15
▶ RunDiffusion reference: bit.ly/RunDiffusionLevendeStreg
▶ posemy.art/
▶ Line to import model: aria2c -x 16
▶ Guidelines video for Dreambooth training: • Automatic1111 Stable D...
-------------------------------
My setup
Macbook Pro M1 Max: amzn.to/3Znd44l
Wacom Cintiq 32 inch (couldn't find the 32 inch - so the link is for 27 inches): amzn.to/40ubLSl
Atem Mini Pro: amzn.to/3nAe9s4
HAG Capisco Adjustable Standing Desk Chair: amzn.to/3Hs8k7t
Brydge dock: amzn.to/3Zr5Kob
Stream Deck XL: amzn.to/3Zun5N9
Razer Tartarus v2: amzn.to/3TV239a
Logitech MX master: amzn.to/3TSFK43
-------------------------------
00:00:14 - And then we are going to do LoRA training and CKPT training in Stable Diffusion - in UNDER 5 minutes.
00:00:56 - So in Midjourney, I am creating a comic book character - by using a design sheet and creating images of my character looking like Miley Cyrus.
00:01:39 - When you create comic book characters, it's essential to create some characteristics
for your character so they are easily recognized throughout your frames or panels.
00:02:37 - So if you do a short comic around 16 to 20 panels, this will take too much time. It's easier and faster to draw the panels. But if you're doing a comic of 200 panels - this is what you want to use. Also if you cannot draw, this is also the process to use.
00:03:51 - To create a good LoRA model you need a least 18 images from all angles, close-up shots, medium shots of the face and body, and long shots too. Make them 512x512 pixels.
00:04:21 - 1 to 1 aspect ratio on the images we train on.
00:05:02 - So the dream booth service on Dreamlook.ai is what we will be focusing on today.
00:07:25 - RunDiffusion is a cloud service that runs Stable Diffusion. We'll import our models in there to continue promoting.
00:09:30 - To create good poses go to Posemy. art. to use with ControlNet in Stable Diffusion.
00:16:11 - How to use the images in ControlNet to control the character pose.
Watch Our Top Videos:⏬
✅ Midjourney - This will change how you write prompts! (SECRET RECIPE)
👉👉 • Midjourney - This will...
✅ Midjourney - comic characters + the SECRET of fixing hands and eyes!
👉👉 • Midjourney - comic cha...
✅ THIS IS CRAZY!!! The perfect poses made easy! Multi ControlNet, PoseX, ......
👉👉 • THIS IS CRAZY!!! The p...
✅ This changes everything! Control colors, people, and poses - Multi ControlNet ......
👉👉 • This changes everythin...
✅ ChatGPT 4 - You're doing it WRONG. Unleash the power of AI and Midjourney 5!
👉👉 • ChatGPT 4 - You're doi...
#LevendeStreg #StableDiffusion #CreateComicBookCharacters #LoRATraining #LoRATraininginUNDER5MINUTES #LoRATraininginStableDiffusion #ComicBookCharacters
After trying a few different ways locally to make LORAs and getting a lot of errors, this worked so well for me. Thank you so much!
Thank you kindly for your comment, and i'm glad it worked for you! Thanks for watching!🙏
Nice tutorial, thank you. Bu what about training Lora on transparent PNG pictures ? What SD will think then ? Also can I train on SD 1.5 model, but with 768,768 resolution ?
Thank you kindly for watching and commenting. As for transparency there’s is a fix for that I think, but as such SD cannot understand transparency as of yet. There are some solutions and extensions where you can put color in and describe what you want SD to create for each color. But not transparency. And for the resolution of 768 ox - as far as I could see Dreamlook can only train on 512. But of course if you train on your own local SD or RunDiffusion then Training works just fine on larger images. They just take a little longer to render.
Hi ! You made my day! Since the dreambooths colabs were broken I wasnt able to train ckpt models until now, thank you so much, I thought that I wouldnt be able to train custom models anymore
I'm so glad I could help. And thanks for the comment and for watching. 🙌
Awesome!
Thank you kindly, Ex UNO! 🙌
good video and tutorial, but I didn't understand the aria2c command or find any details about it in your description. I think that part would be easier on a regular stable diffusion setup as opposed to using the run diffusion service with the creator club upgrade.
I agree. It would be a little easier on a local install. But really, it works very well on Runpod and RunDiffusion. So no trouble at all. It's just a line of command. And the power of RunDiffusion (and their huge graphics cards) is so much faster and easier. The most important part for me with that solution is that I don't need to maintain any of the code. It's someone else's headache. And there is so much maintenance and work with keeping the code up to date and importing and adding new extensions. So yeah, I'm very happy to use a cloud solution for that.
this is soooo helpful. thank you
You’re very welcome. So glad you liked it!🙌
Thanks for your content, its amazing
Thank you kindly for watching and commenting. It means so much to me.
In posemyart, are we able to manipulate the pose by red/blue dots to move the joints so it changes the pose?
Yes, that is exactly what you are able to in Posemyart. Hope it works well for you.🙌
Great video! It helped me alot!
So glad to hear it. Thanks for watching and commenting 🙌
pretty cool thanks!
Thank you for watching and commenting. It means a lot to me🙌
So much appreciate the energy you put into your vids! That makes learning that much more fun, thank you!
Thank you for your kind comment. You're very welcome. So glad you liked it!🙌
Nice tutorial. Can I draw my own character in many poses and use those images for the Lora and model train?
Yes that’s exactly what you can do🙌
Rarely do I come across a fellow Dane who speak English so proficiently, almost no accent at all! Also, thanks for the stellar video, I really want to train a character LORA. I used SD Dreambooth about 10 months ago, in a rented google server. Took 1 hour, and results were not good.. I run SD locally, but last time I tried, dreambooth was too heavy for my GTX 1080
Thank you kindly🕺- Yeah dreambooth is heavy now. You can try out RunDiffusion (it's a cloud service). I have a promo code to their service for 15% off on the first month of Creator's Club. It really works well. Promo Code: levendestreg15. Hope it works out. 🙌
@@LevendeStreg Heh okay, thanks :)
Ngl that helped me sooo much
I’m so glad you found it useful. Thanks for watching and commenting!
tried the checkpoint are amazing but the extract lora doesn't work at all sadly
Glad to hear the CKPT works well. With LoRAs it’s more difficult- it feels like you need to test them. They’re more sort of a style, I feel. So they’re great as addition for a CKPT…
I have uploaded all the models,
I am able to select my Stable Diffusion checkpoint,
But how do i select my Lora model? I have uploaded it. but cannot see how to activate it like you
For anyone else having trouble, at 8:07 You need to type in " aria2c -x16 (link) " You will need to paste your Run diffusion Models LINK
Hope this helps
Yup - aria2c -x16 (link) is correct!
If I try to make a cloth lora, should I do the same as in the video? Is there any difference?
You mean if you want to make a LoRA with clothing? Yeah, then you would dress the person in specific clothing and have pictures from different angles. If you only want ONE piece of garment (like a hat) on different people - then you need lots of images with different people wearing that very hat - from all angles. Hope it helps.
Mike deodato is an amazing artist. Frank Cho also has nice colours ;]
Yeessss! They are both great! 🙌
Great video! Very helpful, thank you!
Wow. Thank you so much Louis. I really appreciate it 🙌🔥
What is that ARIA 2c you are talking about?
It's the old line (that you at the time had to use, to import Models into RunDiffusion). Now they've built in a button to just import models.
Excellent (Y)
Thank you so much for watching and commenting. I really appreciate it.🙌
Thank you so much! This is wonderful knowledge! So helpful tutorial!
Thank you kindly. So glad you liked it🙌
Can m1 pro sd use lora though?
I don't use my local install anymore. I breaks all the time - and no - dreambooth doesn't work with M1 chip or M2 chip unfortunately. That's why I use a cloud service like RunDiffusion for running Stable Diffusion. Be sure to use the promo code levendestreg15 to get 15% off of that, if you chose to use that solution.
how to train/insert/inject own photo of clothes,costume,dress
Thanks for asking. I’m gonna show that in an upcoming video - basically it takes a lot of inpainting and correcting with Photoshop and stable diffusion…
so it's bad to make a transparent background for the dataset images. (i need the backgrounds.
Great question. It depends on what you wish to achieve. You cannot have transparent backgrounds. But you can have an all white or black background for instance - say if you're trying to create product shots of a product. And that would work great. But no, you cannot have transparent backgrounds. You might want to try adobe photoshop or shutterstock to get rid of the background AFTER you create your image with AI. But you cannot create images without background / transparency yet.
Thanks for the fast reply... 👍
You're welcome. Hope it works,@@talkdirty7777 🙌
Shout out for video idea.
"Complete newbie guide to sd & 1111"
All words & tech jargon explained.
What resources u need to download
I dont think theres anything like this on YT.
Fingers crossed ;)
Hehe - thank you for the idea, the reason why there is not a video like that is that I’d good difficult to make in just one video. That would be a course in it self… but I am looking into doing a course🙌
@@LevendeStreg hey hey hey. Good idea. Sounds epic ;)
@@armondtanz Thank you. Glad you like the idea of a course🕺
ty
You're very welcome! Thanks for watching and commenting!🙌
Excuse me what is the site in the minute 5:03
That's dreamlook.ai (see the link in the description). Works really well.
thanks peso pluma
Hi @hypergaming3057 - what does "Peso Pluma" mean?
So many refer to this video
I'm so glad to hear it. Thanks for watching and commenting. 🙌
👋
Thank you Louis. So nice of you to drop by 🙌🙏
Is this still best solution you found for comic book characters? anything changed over past 6 months? Please let us know if you are using something else now or any new flow. Thanks.
This does litterally nothing. I generate an image, save the seed and all settings, same prompt. I turn on the lora (with they keyword) and regenerate, exact same image is generated. This does nothing at all
Sorry it doesn't work. It's been some time since I tested LoRAs. Please note this is an old video from many month back. Hope you get it to work.🙌
so we have to pay for many services
Thank you for your Comment. Sorry - but AI is really hard to learn and manage. Not to mention the expense if you have to buy a huge Nvidia graphics card. So instead of using the GPU on my computer (which doesn't really work for Mac users) I recommend using cloud tech. It's much faster, simpler and thus also better to fit my professional workflow. 🙌
Ok, that video was 16:49 long. During the first 5 min I didnt lear a thing.
THank you for watching and commenting. It is an older video. And the service has improved greatly since I made this video.
5 minute to tune bakchodi me laga diye
Hi there - not sure I understood your comment. I’m not very proficient in Hindi… 😉
Stupid pay sites.
I don’t agree. I’m actually very pleased with Dreamlook.ai - just like I like to use Adobe and Pay for their products. Thanks for watching and commenting.
Why do all AI channels feel like advertisements for cloud services?
Well you can do that on your home PC too if its good enough. But i don't tell you how that works. I tell you how to use a cloud service.
True - but I have a MacBook Pro Mac - and I cannot use a local install with all the extensions I want - because the M1 and M2 chips can’t run it - and you cannot use an external graphics card with that on new Macs. So that leaves me with cloud solutions!
@The Padded Cell Thanks for defending my honor. I really appreciate it🫶🙏
Genuinely funny that a generation of designers still think a woman with his haircut is "edgy"
Hahaha… I’m not sure it’s edgy. But as AI has been trained on millions of images - it’s safe to say that women (according to training material) look a certain way. Think long hair and makeup and NOT a lot of clothing and TIGHT clothes. So if you want to create something different than that - you need to specify it in the prompts. Otherwise when you prompt for a woman, you will get a very feminine person with dress, long hair, small tops and so on. So yes, the women I often prompt for goes against the “normal” and “trained” version of a woman.
very bad explanation at the end!
Well thank you for watching to the end, I appreciate it. Why do you think it was a bad explanation? Cheers 🙌
Nuh uh
kool!
Thank you kindly. So glad you watched and liked🙌