MorphAny3D: Unleashing the Power of Structured Latent in 3D Morphing
Xiaokun Sun, Zeyu Cai, Hao Tang, Ying Tai, Jian Yang, Zhenyu Zhang
2026-01-05
Summary
This paper introduces a new technique called MorphAny3D for creating smooth and realistic 3D morphs, which are transitions between 3D shapes.
What's the problem?
Creating good 3D morphs is hard because it's difficult to make the changes look natural and consistent over time, especially when you're morphing between very different types of objects, like a car turning into a chair. Existing methods often produce distorted or unrealistic results.
What's the solution?
The researchers developed MorphAny3D, which doesn't require any training data. It works by cleverly combining information from the starting and ending shapes using something called 'Structured Latent' representations. They created two new attention mechanisms: one to ensure the structure of the objects stays coherent during the morph, and another to make the morphing sequence look smooth from frame to frame. They also added a step to correct for differences in how the objects are oriented.
Why it matters?
This work is important because it allows for the creation of high-quality 3D morphs without needing a lot of data or complex training. This opens up possibilities for advanced applications like changing the style of a 3D model or creating seamless transitions between different 3D objects, and it can be used with other 3D generation techniques.
Abstract
3D morphing remains challenging due to the difficulty of generating semantically consistent and temporally smooth deformations, especially across categories. We present MorphAny3D, a training-free framework that leverages Structured Latent (SLAT) representations for high-quality 3D morphing. Our key insight is that intelligently blending source and target SLAT features within the attention mechanisms of 3D generators naturally produces plausible morphing sequences. To this end, we introduce Morphing Cross-Attention (MCA), which fuses source and target information for structural coherence, and Temporal-Fused Self-Attention (TFSA), which enhances temporal consistency by incorporating features from preceding frames. An orientation correction strategy further mitigates the pose ambiguity within the morphing steps. Extensive experiments show that our method generates state-of-the-art morphing sequences, even for challenging cross-category cases. MorphAny3D further supports advanced applications such as decoupled morphing and 3D style transfer, and can be generalized to other SLAT-based generative models. Project page: https://xiaokunsun.github.io/MorphAny3D.github.io/.