SDXL LoRA/LyCORIS works best on its original training model. You may need to reduce weight when using it with other models or additional LoRAs.
LyCORIS Combination Recommendation: Try using four of my LyCORIS models with the following weights: 0.4, 0.3, 0.2, 0.2 (or any combination with a total weight around 1.0).
I will release multiple versions for a few popular models. Feel free to request for artist or model.
Version naming convention: MajorVersion.MinorVersion [BaseModel]
MajorVesion: For big updates that can apply to any style LoRA. For example, training parameter update.
MinorVersion: For small updates that only apply to this LoRA. For example, epoch selection, removing a certain image from dataset.
BaseModel: SDXL LoRA/LyCORIS works best on the model that it was trained on.
Change notes:
v2.233->v2.233-2: Switching to a new epoch selecting strategy: reducing overfitting and improving hands.
Description
FAQ
Comments (7)
I can see new version for ponyxl use more data from a little bit early ages, I think it is also great. But I just want to remind you that since this artist change his style in recent years, old images can be harmful for this Lora.
Thank you for your suggestions. I will try removing old images. I still have many ideas to test (including training a character LoRA). Likewise, I'm not going to do it soon.
@silvermoong Hello fellow AI enjoyer, wich program do you use for training, i pretend to create obscure artists lora, that have very good styles, but i can't downgrade my python otherwise it could causes tecnical issues, so...
@silvermoong after further testing, I think for this lora, I found some overfitting issue. For example, if you use some rare clothes tag or hair color tag, model may automatically make it like one of those characters in the doujinshi/game CG of the artist. I think the correlation between some tag and characters is learned instead of tag and corresponding features. (I did not test your previous Lora for this artist due to short of time, so I do not how this come.)
@os12z I found an overfitting issue in my epoch selection strategy yesterday. I'm about 10-20 more epochs than the edge of overfitting. (My goal is to pick the highest epoch without serious overfitting, as users could reduce LoRA weights or use on a non-PonyDiffusionV6 model.)
Could you please share me an example of the rare cloths and hair color tag?
@dolranaegsa428 Use VENV for your training environment.
Regarding training program, I use https://github.com/Akegarasu/lora-scripts
@silvermoong Sorry to reply late, I am too busy on my own job. And I am not quite sure, I think like sailor suit, tiara, etc. These tags may cause the character harder to change clothes, and the word, nun, may cause character to have more blue hair. In my personal experience, I met such kind of cases. I think maybe I am a little bit picky on this lora since I see a lot of version made by you and have some kind of... deja vu?