Text-to-hologram: hologram generation from natural language

Computer-Generated Holography (CGH) enables high-fidelity three-dimensional (3D) visualization but faces challenges due to the complexity of inverse wavefront calculations and the substantial data volume required for holographic content transmission. To overcome these challenges, we propose the Text-to-Hologram (Text2Holo) task, which enables users to generate and create holograms solely through natural language input—without requiring any programming expertise or knowledge of computational imaging. By leveraging generative models, Text2Holo eliminates the need for predefined image inputs, significantly enhances accessibility, and democratizes holographic content creation. Text2Holo enhances data efficiency by compressing holographic information into a text-based format. Experimental results validate the effectiveness of our approach, achieving a peak signal-to-noise ratio (PSNR) of 31.02 dB and a structural similarity index (SSIM) of 0.84. By introducing natural language as a direct input modality for hologram generation, Text2Holo broadens the applicability of CGH, paving the way for next-generation immersive holographic displays and enhanced AR/VR experiences.