“It will ultimately be a multimodal model, trained on text, images, video, and other types of input – and will integrate with our existing AI creation tools,” according to Roblox.