Meet ReCo: An AI Extension for Diffusion Models to Enable Region Control - MarkTechPost

Large-scale text-to-image models, looking at you Stable Diffusion, have dominated the machine learning space in recent months. They have shown extraordinary generation performance in different settings and provided us with visuals that we never thought were possible before. Text-to-image generation models try to generate realistic images with an input text prompt describing what they should look like. For example, if you ask it to generate "Homer Simpson Walking on the Moon," you would probably get a pleasant-looking image with mostly correct details. This huge success of generation models in recent years is mainly thanks to the large-scale datasets and models used.