Wow! Vision banana confirms that image generators are great generalist vision learners. Pretraining alone gets you zero shot segmentation, depth, normals etc – while beating out specialist models. If you teach a model to draw – you also teach it to see! pic.twitter.com/FGsyzxlrmW
— Bilawal Sidhu (@bilawalsidhu) 23 avril 2026
Wow! Vision banana confirms that image generators are great generalist vision learners. Pretraining alone gets you zero shot segmentation, depth, normals etc – while beating out specialist models. If you teach a model to draw – you also teach it to see!