Dreambooth gives better results than LORA. Especially if you can train on Dreambooth with EMA enabled, which you need around 20GB of VRAM for. Though, I agree

Dreambooth gives better results than LORA. Especially if you can train on Dreambooth with EMA enabled, which you need around 20GB of VRAM for. Though, I agree with you that the real product here is the pipeline. The parent comment is also correct though in that the majority of people aren't in a place where they can train their own models yet despite it being possible for them. This product is taking advantage of the current arbitrage opportunity.