FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation

2026-06-23

AI Summary

Existing image-to-3D Gaussian splatting methods are hindered by sparse voxel representations and cross-modal alignment bottlenecks, limiting their ability to preserve high-frequency details. This work proposes FLUX3D, a novel framework that introduces Diffusion-Aligned Structured Latent variables (DA-SLAT) and a sparse structure-aware diffusion architecture comprising a Sparse Multi-scale Diffusion Transformer (SMDiT) and Multi-modal Axial RoPE (MARoPE), enabling geometry-agnostic precise 2D–3D alignment. By leveraging a decoder-only design, sparse voxel representation, diffusion-based Transformer, and multimodal rotational positional encoding, FLUX3D achieves significantly higher appearance fidelity than current state-of-the-art methods while efficiently generating high-fidelity 3D Gaussian splatting assets.

Abstract

Sparse voxel representation has emerged as a scalable foundation for image-to-3D Gaussian Splatting (3DGS) generation, yet current methods struggle to preserve high-frequency visual details of input images due to two structural bottlenecks. First, they adopt discriminative 2D features optimized for semantic abstraction to construct sparse voxel latents, which suppress reconstructive cues and induce a representation bottleneck. Second, in the generation stage, standard diffusion transformers lack effective mechanisms to align dense 2D image tokens with sparse 3D voxel latents, resulting in a cross-modal correspondence bottleneck. To address these issues, we propose FLUX3D, a scalable image-to-3DGS framework that boosts both representation learning and cross-modal alignment during generation. We first revisit 2D feature selection for sparse-voxel-based 3D representation learning, propose Diffusion-Aligned Structured Latents (DA-SLAT) and couple it with a decoder-only architecture to improve 3DGS reconstruction fidelity. We also design a sparse-structure-aware diffusion framework, which integrates the Sparse-structure Multimodal Diffusion Transformer (SMDiT) and Modal-Aware Rotary Positional Embedding (MARoPE) to achieve geometry-agnostic 2D-3D alignment. Extensive benchmark experiments demonstrate that FLUX3D yields substantial improvements in appearance fidelity and significantly outperforms all state-of-the-art (SOTA) methods in generating high-quality 3DGS assets.

Problem

• 3D Gaussian Splatting
• sparse voxel representation
• high-frequency details
• cross-modal alignment
• image-to-3D generation

Innovation

• Diffusion-Aligned Structured Latents
• Sparse Voxel Representation
• Multimodal Diffusion Transformer
• 3D Gaussian Splatting
• Cross-Modal Alignment