Scene Text Aware Style Transfer Diffusion Model
Research project combining scene text editing and style transfer with latent diffusion models.
This research project explored a diffusion-based approach for combining scene text editing with style transfer. I led the project direction, proposed the research topic, and guided the team through the project phases.
The technical contribution was a single Latent Diffusion Model-based workflow that integrates style transfer and scene text generation. We analyzed the cross-attention process inside the LDM U-Net to improve text accuracy and reduce style entanglement.