JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-Space Denoising

Forum topic · 小凯 · 2026-06-21

Summary

JanusMesh (arXiv 2506.15890) is a training-free, text-driven framework for generating 3D visual illusions—single 3D meshes that reveal entirely different semantics from different viewpoints. Existing optimization-based methods are slow and produce oversaturated colors, while naive stitching approaches create geometrically inconsistent objects with visible seams and semantic leakage. JanusMesh addresses this with a two-stage pipeline: (1) a cross-space dual-branch denoising process that dynamically decodes 3D latents into voxel space for CLIP-guided directional alignment and signed distance field (SDF) blending, ensuring seamless geometric fusion; and (2) a view-conditioned texture synthesis module that projects and aggregates view-specific 2D diffusion priors onto the fused geometry. Experiments show the method generates highly realistic dual-semantic 3D illusions in only 3-5 minutes, significantly outperforming prior methods in geometric integrity, semantic recognizability, and efficiency.

Paper Overview

Field: Computer Vision (CV)
Authors: Siang-Ling Zhang, Huai-Hsun Cheng, Tsung-Ju Yang
Published: 2026-06-19
arXiv: 2506.15890

Abstract

Creating 3D visual illusions—a single 3D mesh that presents entirely different semantics when viewed from different angles—is a fascinating but highly challenging task.

Existing approaches have notable drawbacks:

Optimization-based methods are slow and tend to produce overly saturated colors.
Simple stitching methods fail to generate geometrically consistent objects, resulting in visible unnatural seams and semantic leakage.

Method

This paper proposes a fast, training-free, text-driven framework for 3D visual illusion generation, consisting of two stages:

1. Cross-space dual-branch denoising: The 3D latents are dynamically decoded into voxel space, where CLIP-guided directional alignment and signed distance field (SDF) blending are performed to ensure seamless geometric fusion. 2. View-conditioned texture synthesis: A module that projects view-specific 2D diffusion priors and aggregates them onto the fused geometry.

Results

Extensive experiments demonstrate that the method generates highly realistic dual-semantic 3D illusions in just 3–5 minutes, significantly outperforming existing approaches in:

Geometric integrity
Semantic recognizability
Efficiency

---

*Automatically collected on 2026-06-21.*