Requests/ideas
If you have any requests or ideas for LoRAs, drop them here and I will put them on the list for the future. (No promises though.)
I want to know how are you training these Loras like some recommendations on the parameters and whats the amount of dataset you need to generate good quality Loras?
I'm using OneTrainer, for no other reason than that's what I'm familiar with. When training your first LoRA, use their preset just to see what that does. But after that I would start fiddling with the learning rate, making it higher.
As for datasets, make sure you properly caption them. Other than that it's hard to say. I've made great LoRAs with just 13 images and others with about 50. It just varies.
How do you inference these Loras after finetuning. I get error while i am using load_lora_weights. if you can kindly provide the test code for Lora?
Also an idea of trying to train for realism Lora and share your findings.
I just use a standard ComfyUI workflow for Chroma, nothing really fancy. You might need to update.
I don't train for realism at all, I'm simply not interested in doing so. (My 3D LoRA is probably the closest I will come.)
Go to Lodestones discord if you want to talk to the real experts on Chroma: http://discord.gg/SQVcWVbqKx