Similar training parameters to v1 except I used constant instead of cosine_with_restarts and stopped at epoch 17 - The resulting LoRA produces images more inline with my intention. Seems easier to get results with simple prompts.
Similar training parameters to v1 except I used constant instead of cosine_with_restarts and stopped at epoch 17 - The resulting LoRA produces images more inline with my intention. Seems easier to get results with simple prompts.