Multimodal Image Colorization: Quantifying the Impact of Text-Conditioned Guidance on Grayscale-to-Color Translation

This study evaluates the impact of text conditioning on grayscale-to-color image translation using two architectures: U-Net and Stable Diffusion 1.5. The introduction of CLIP text conditioning resulted in significant improvements in performance metrics, with U-Net showing a 5.6% increase in PSNR and a 36.6% increase in colorfulness, while Stable Diffusion 1.5 achieved a 5.8% increase in PSNR. These findings underscore the effectiveness of integrating text guidance in enhancing the quality of automated colorization, which is critical for applications in historical restoration and medical imaging.