Lumiere Google Research's new text-to-video diffusion model can also handle inpainting, cinemagraphs, and converting still images to video #