Closed Wenaka2004 closed 2 weeks ago
The dev branch will work with v-prediction SDXL model. NoobAI_XL may assume zero terminal snr. If so, please add --zero_terminal_snr
option.
Thank you for your response and the solution! I’ll give it a try.
Training has been successful, thanks.
does this work on the schedule free branch?
Please add support for training LoRA models with v-prediction SDXL models. The v-prediction version of the NoobAI_XL model has shown excellent performance, and adding this feature would enable further training based on it.
Currently, training LoRA on v-prediction models does not throw errors but results in empty images during generation.