I am exploring ways to integrate some custom loss functions into the Stable Diffusion model to impose specific constraints during training. Given my limited computational resources, retraining the model from scratch is not a feasible option.
I am considering the use of LoRA for fine-tuning the model with the new loss function. However, I am unsure if LoRA has the capability to effectively incorporate such changes.
Would LoRA be an appropriate tool for fine-tuning with custom loss functions, or should I look into other techniques such as DreamBooth or Custom Diffusion approaches to achieve my goal?
Thanks for your help.