10,000x smaller & 25x faster? My old Google teammates & their collaborators, who changed the generative game last year by enabling custom model training, are now proposing to upend things further by enabling training via a single image—and massively faster, too. Check out this thread: