New 1.2B parameter text to video model is out, higher quality than modelscope