This code repository includes research on learning from synthetic image data (mainly images), including three projects: StableRep, Scaling, and SynCLR. These projects explore how to utilize synthetic images generated by text-to-image models for training visual representation models and have achieved very good results.