Re-LAION-Caption 19M
- The paper introduces Re-LAION-Caption 19M as a large-scale dataset that uses a fixed four-part caption template to improve prompt adherence and text-image alignment.
- The dataset is derived from Re-LAION-5B through rigorous filtering steps, retaining 19M high-resolution 1024x1024 image-caption pairs with consistent, structured captions.
- Empirical findings reveal that models fine-tuned with structured captions exhibit slightly better controllability and VQA-based alignment scores compared to using randomly shuffled captions.
1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010^ 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 101010 1010 1010 1010 1010 1010^ 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010^ 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010^ 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010^ 1010 1010 1010 1010 1010 1010 1010 1010^ 1010^ 1010 1010 1010 1010 1010 1010 1010^ 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 101010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 101010 101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 101010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 101010 1010 1010 1010 1010 1010 1010^ 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^( Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 Captions Improve Prompt Adherence in Text-to-Image Models101010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010^ 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010 1010