The article discusses a new concept called ‘prompt-aligned personalization’ for improving the performance of text-to-image models. It addresses the limitations of existing personalization methods, such as compromising the ability to personalize or align complex textual prompts. The authors claim this new approach improves text alignment and enables the creation of images from complex and intricate prompts. Their method keeps the model aligned with a target prompt using an extra score distillation sampling term. The article also provides comparisons of this method with existing techniques.
Publication date: 11 Jan 2024
Project Page: https://prompt-aligned.github.io/
Paper: https://arxiv.org/pdf/2401.06105
