Basically asking for support for models trained on zero-terminal SNR noise schedule and velocity prediction such as the terminus series.
Huggingface link:
https://huggingface.co/ptx0/terminus-xl-gamma-v1
Current version:
https://huggingface.co/ptx0/terminus-xl-gamma-training
Reason? These fully trained from scratch models are significantly cheaper to train, requiring only a single A100 card.
While they outperform base sdxl in a significant amount of prompts, they are currently lacking Lora support, which would really help push them to the next level.
Basically asking for support for models trained on zero-terminal SNR noise schedule and velocity prediction such as the terminus series.
Huggingface link:
https://huggingface.co/ptx0/terminus-xl-gamma-v1
Current version:
https://huggingface.co/ptx0/terminus-xl-gamma-training
Reason? These fully trained from scratch models are significantly cheaper to train, requiring only a single A100 card.
While they outperform base sdxl in a significant amount of prompts, they are currently lacking Lora support, which would really help push them to the next level.