i really appreciate it, i just started using a1111 after using online services and it takes more lea
i really appreciate it, i just started using a1111 after using online services and it takes more learning so thank you
Fyou are welcome
Fswarmui is also amazing to use
Fit is ui of future :d
Hi see. my computer cannot handle any of this, so i have to use something like runpod
Fyep runpod good. massed compute better 
Fwe have them pre installed there
Hcan i just switch everything from runpod to massed?
FYouTubeSECourses
Unlock the power of FLUX LoRA training, even if you're short on GPUs or looking to boost speed and scale! This comprehensive guide takes you from novice to expert, showing you how to use Kohya GUI for creating top-notch FLUX LoRAs in the cloud. We'll cover everything: maximizing quality, optimizing speed, and finding the best deals. With our exc...
Fa lot of things you will learn
Falso i have one trainer tutorial for massed compute
Fit is very good as welll
FYouTubeSECourses
In this tutorial, I am going to show you how to install OneTrainer from scratch on your computer and do a Stable Diffusion SDXL (Full Fine-Tuning 10.3 GB VRAM) and SD 1.5 (Full Fine-Tuning 7GB VRAM) based models training on your computer and also do the same training on a very cheap cloud machine from MassedCompute if you don't have such compute...
Fevery tutorial will teach you something new @Hustlers4Ever
Hexcellent thank you
MHey there! Sorry to bother you again but I forgot a question the other night… When doing SDXL Dreambooth training you suggest 200 reps, 1 epoch, which in your example here totals 8400 steps. I'm referring to kohya scripts for training. My questions:
1. What happens if you're running more than one GPU, in this case? Do you use 200/[number of gpus] repeats?
2. Would you suggest to modify the main learning rate also in the case of SDXL training, as you suggest with Flux training?
1. What happens if you're running more than one GPU, in this case? Do you use 200/[number of gpus] repeats?
2. Would you suggest to modify the main learning rate also in the case of SDXL training, as you suggest with Flux training?
Fgood question
Fye only way looks like reduce repeat
Fuse same learning rate formula
MI'm getting terrible, awful results with Onetrainer, training for one person
MI have 175 photos of the model, I guess I should slim the dataset for quality, you think?
Fthat would probably work amazing on flux
Ftry to reduce 20 test it
Flater you can expand
MYea my experience is the models go from under to overtrained between epochs
MKohya, OneTrainer, CivitAI/kohya, you name it, all the same
MFlux training is incredibly easy and effective, I'm not sure it's worth for me to keep insisting on SDXL
Fwell i would move to flux
M@Furkan Gözükara SECourses In the latest tutorial, what is the effect of adding an inpaint of your (ohwx) face if you're already the "ohwx man" in the photo?
Fit improves face significantly
Fyou can test
Fsince inpainting is 1024x1024
SHas it been tested on flux training whether captions is way better vs without captions? What about background? Can I just have pictures of me and the backgrounds removed or will that create an awful result?
Ffor person training doesnt help
Fyou can see in research article
Pon runpod A40 vs RTX4090 which better for training lora?
Frtx 4090 faster
Falmost same quality
Fa40 may get just a little bit quality boost from fp16 training
D@Furkan Gözükara SECourses someone seems to have achieved FLUX training on 8GB vram
DDsettings in video description
Didk if you also have, haven't followed FLUX closely yet. But when I last looked at your settings I think it was still > 16 GB, unless you use very slow kohya model splitting technique
SDr Furkan when OneTrainer have full support on flux do you think regularitation images could be usefull again?
Fwe already have config for that
Fuses 7.5 GB
Fi havent tested yet plan to test
Fonetrainer as well
