TOLA One-Step Latent Adaptation for Diffusion-based Text Super-Resolution

TOLA achieves text image super-resolution in a single step by replacing iterative image-text diffusion with a confidence-weighted text conditioning module. This approach prevents error propagation from unreliable OCR predictions during the reconstruction process.