A modified version of the T5-XXL encoder, using a technique inspired by Kaoru8 but with significantly fewer newly added vocabulary items. It was originally designed to improve adherence to Danbooru-style tags. However, in my initial experiments using SD-Script to train a T5 LoRA, the performance was not very satisfactory. Therefore, this may require a long-term fine-tuning process.
- Downloads last month
- 4
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
๐
Ask for provider support