Residual Decoder Adapter: ID-Preserving Tokenizer Adaption for Autoregressive Text Rendering

ArXi:2606.01911v1 Announce Type: new Visual Autoregressive (AR) models generate images by predicting discrete tokens that are decoded by a visual tokenizer. Despite nstrating strong overall image generation ability, they still underperform on text rendering with blur strokes and disrupt letter shapes. In this work, we trace this limitation to the visual tokenizer, which struggles to reconstruct fine-grained detail. Improving the tokenizer is straightforward but expensive, as it necessitates re.