metadata
base_model:
- stable-diffusion-v1-5/stable-diffusion-v1-5
- zer0int/LongCLIP-GmP-ViT-L-14
sdLL model - SD1.5 plus LongCLIP
The long-term purpose of this is currently for experimentation leading up to a better version of our XLLSd model. However, in theory, it could be used as a standalone model, or as a basis for other, fancier models.
Basically, this is SD1.5 plus LongCLIP smooshed togther with zero extra training. The reason to create this, is to have base SD1.5 with a CLIP that can handle 248 tokens instead of only 75
This should in theory make things like natural language prompting for SD much more practical, without having to got all the way to things like T5 text encoders, etc.