LORA DATASET MAKER
Can anyone tell me what would happen if I insert ton of images in the dataset for LORA Creation without giving them captions?
Recently got a good example of what happens:
The dataset is really visually busy in the background but that isn't tagged properly. So that is incorporated into all images.
Any part that is "consistent" for all images without captions will be incorporated into the LoRA.
That is the same way you make good character LoRA. By leaving out the omnipresent prompts (hairstyle, eyecolor, bodyshape etc etc) these will be included in the LoRA/activation prompt.
As dolirama said: If this consistency isn't there it can end up a total mess.