LIKE TO SEE YOUR RESULTS!
I recommend trying CFG=5. not necessary, but might make it better. I am writing about it, because many people use a higher parameter.
Description
1- upload dataset.
2- made a model based on the settings of other creators. Added tags during training. Taught everything also on anything 4.5
FAQ
Comments (9)
how many images did you train with i think so aomb doesn't support very well i try it with my mix it's pretty awesome thanks for your lora i hope there are more lora
Ah, I forgot to edit the HD description. It was on the old model. Everything should be fine now
Hello if its possible, can you make SDXL version of this?
Yes, I was thinking about doing it. Lora on SDXL trains very strangely. it's hard to trace the golden mean. Maybe the results of ishida akira will be more clear which stage of training is better. It wasn’t clear at all on sakimichan, I had to choose at random.
I did it! To my surprise, I'm happy with the result.
@Simisir How about the old one? iirc pony model doesnt work well with other model
@Zet69 I don't think it's worth updating the old one. sdxl will not learn quite correctly. There are nuances there that I don't really want to bother with. I do it first and foremost for myself. The response from other people is negligible, so there is no desire to do anything additional.
@Zet69 again, you say one thing, but you didn’t even show your work in the review. I don't think this is very good behavior...
@Simisir Chill out, I am relatively new here. There I already reviewed it