A new framework called OmniManim enhances the generation of educational animations by addressing visual defects through render-feedback-aware code generation. It employs a visual planning module to improve render quality and efficiency, outperforming existing models in key metrics.