Tsinghua University Professor Liu Yongjin's research group has achieved SOTA results in the text-to-3D field with their TICD model, where multi-view consistency priors enhance the quality of 3D models. The TICD method demonstrates comprehensive superiority on the T3Bench dataset, achieving the best results for single object, single object with background, and multi-object prompts. By incorporating NeRF supervisory signals conditioned on text and images with multi-view images, TICD addresses the limitations of pre-trained diffusion models. The TICD workflow includes sampling orthogonal camera viewpoints and NeRF rendering reference views.