JanusMesh Framework Achieves Fast, Zero-Shot Generation of 3D Visual Illusions

Researchers have developed JanusMesh, a fast and training-free framework for creating 3D visual illusions that reveal distinct semantics from different viewing angles. Existing methods are often slow and prone to producing unwanted artifacts, but JanusMesh employs a two-stage process. The first stage involves a cross-space dual-branch denoising method that aligns 3D latents with CLIP-guided orientations, ensuring smooth geometric fusion. The second stage utilizes a view-conditioned texture synthesis module that integrates view-specific 2D diffusion priors onto the geometry. Experimental results show that JanusMesh generates realistic dual-semantic 3D illusions in just 3-5 minutes, outperforming previous techniques in geometric integrity and efficiency.