Home Artists Posts Import Register

Downloads

Content

I realized for myself that it’s time to stop experimenting with settings (although I’ll continue anyway), making a dozen versions similar to each other in order to achieve the best result, this is certainly good, but it takes a lot of time and most often it’s just not worth it, the desire to make the “best of the best" will ultimately destroy me, plus there is not much sense from this, because people simply do not read the description of how to improve their images with a banal hires fix, upload images in 512x512 resolution with a minimum number of steps and put 1 - 3 stars, why am I trying? Don’t think that I’m complaining, I know that for 1 person who doesn’t want to understand and just puts 1 star, there are 99 guys who understand everything or want to learn, but it just seems to me that civitai becomes too toxic with the arrival of a huge number of people, this is inevitable and I just need to put up with it and keep working.

Why so many versions? Essentially you only need two from my point of view, 1.3RW and 1.6R

  • 1.4RW and 1.5RW - experiments (unsuccessful from my point of view, drilling forums, I read "experts" on advice from lora, all the advice turned out to be nonsense and what I worked out a long time ago almost flawlessly works, and these versions are not very successful, albeit with its advantages in some rare moments). P.S. patreon does not allow uploading more than 199mb files, so version 1.5 will not be able to attach, I will add google drive
  • All versions - https://drive.google.com/drive/folders/1PO0PKdrSjvMGN5oRwFTCAELocIsgZ4Hz?usp=sharing
  • 1.3RW - the most successful version with new settings based on "warmap" with correctly selected numbers, the version in my opinion is often the best, but again not always.
  • 1.6R - old settings version, just old settings used in 1.2, why did I reinvent the wheel if it only has one wheel? I don't know myself, don't ask.

Oh yeah, actually besides the increased number of characters from the 90s, are there any changes? Yes, I tried to use multiple faces, as I expected, more than two characters are simply unrealistic to count if there are not 100 arts with these characters, so that Lora "learns" how to read them, but the attempt is not torture, now Lora can create a good semi collage random characters in very good quality.

In fact, I understood one thing - you don’t need to listen to anyone and try to jump above your head, the more art, the worse the quality, this is inevitable, you can only negate the drop in quality, but not remove it completely. If I want to make a huge 90s-style lora with a huge number of characters, then I need to put up with a slight drop in quality, or I will go crazy from constantly reworking something that already works at its maximum and does not need to be changed.

Perhaps I need to relax with simple loras for 1-2 characters, or some kind of feature, if there are ideas what I should create, I will be glad to hear!

Files

Comments

Anonymous

Which do you like more, the 1.6R or 1.3RW

OneRing

actually a very difficult question, 1.6R was made in the same way as 1.2R, on the whole it’s not bad and very diverse, but 1.3RW is very good in the details of exactly those arts that I laid for it, while practically without loss in quality, everything is so but there is volume, shadows, good exposure, to put it bluntly, I will give preference to 1.3RW

Anonymous

Do you know if this is Compatible with InvokeAI's Diffusers?

OneRing

I do not think that there is a fundamental difference, so I think that it is compatible