L-MAGIC by Intel

L-MAGIC is a novel method presented at CVPR 2024 by Intel Labs for generating coherent 360-degree panoramic scenes from a single input image or text. It leverages large language models for guidance while diffusing multiple coherent views, using pre-trained diffusion and language models without fine-tuning (zero-shot). The output quality is enhanced by super-resolution and multi-view fusion. It accepts various input modalities including text, depth maps, sketches, and colored scripts, and can enable 3D point cloud generation and dynamic scene exploration.