Actually, LeoAI doesn't even need to have its own image generation pipeline, it could use normal Stable Diffusion inference, and most people will be finfine with that