Text-Printed Image: Bridging the Image-Text Modality Gap for Text-centric Training of Large Vision-Language Models

TL;DR

• Printed text becomes low-cost VLM training data: TPI renders textual descriptions directly onto plain white images.
• Text rendering bridges the image-text modality gap: Provides visual training inputs without collecting real images.
• TPI outperforms diffusion-generated training images: Preserves semantic accuracy and improves results across multiple benchmarks.

Futa Waseda

Project Assistant Professor