ASD is a lifelong disorder and affects subjects’ behavior, the early detection of ASD is required to provide timely treatment. But diagnosing this disorder is challenging for qualified professionals, as there is not any well-established test for autism. Our research proposed an ASD detection model from facial images using transfer learning. Carried out the findings from image based analysis to construct a LSTM model to train on video dataset. Video dataset was self curated across streaming platforms like YouTube, containing both ASD and TD children