Welcome to Anime Natural Language XL , a NEW finetune model trained by natural language.
This is a Anime style model, which can also be said to be an base model on SDXL1.0 training.
This model has changed the training method of finetuning by using natural language training, which can achieve more stable and controllable anime effects and natural language responses.
The dataset contains 60,000 images and five major categories, and the model is capable of achieving diverse anime effects.
We would like to thank Yidhar and Keruchis for their assistance in the training process.
We hope that this model will have the potential to serve as a base for anime-style images, and we will continue to improve it in the future.
Description
FAQ
Comments (6)
I hope you tuned this with SDXL bucketing. SDXL on civitAI have been plagued with 256x1024 bucketing thats been destroying models.
nontheless good work for making a proper natural text finetune !
Did you tune the text encoder by chance?
yep, SDXL bucketing is nonideal so I didn't use it, lol
@L_A_X I'm curious why. the main reason SDXL is good for was it's bucketing. you are able to go above 1024x1024 with ease. This is like finetuning on NAI at 896. There's no parent data in those resolutions above 768 in NAI therefore you have anatomy issues unless u train with a big dataset. but in SDXL if you're pulling a finetune u can easily go over 1024. why limit yourself to 256x1024 bucketing? there's no parent data if you go below 512 in the first place.
Why I can't get the same picture showing above with the same params?
大佬， need 异格杰西卡, plz
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.