Target-Balanced Score Distillation

TL;DR

This paper introduces Target-Balanced Score Distillation (TBSD), a novel method that balances texture realism and shape accuracy in 3D asset generation by adaptively managing negative prompts, outperforming existing techniques.

Contribution

The paper reveals the fundamental role of Target Negative Prompts in the trade-off between texture and shape, and proposes TBSD as an adaptive multi-objective approach to improve 3D asset quality.

Findings

TBSD outperforms state-of-the-art methods in texture fidelity.

TBSD achieves high geometric accuracy in generated 3D assets.

Systematic analysis clarifies the impact of negative prompts on 3D generation.

Abstract

Score Distillation Sampling (SDS) enables 3D asset generation by distilling priors from pretrained 2D text-to-image diffusion models, but vanilla SDS suffers from over-saturation and over-smoothing. To mitigate this issue, recent variants have incorporated negative prompts. However, these methods face a critical trade-off: limited texture optimization, or significant texture gains with shape distortion. In this work, we first conduct a systematic analysis and reveal that this trade-off is fundamentally governed by the utilization of the negative prompts, where Target Negative Prompts (TNP) that embed target information in the negative prompts dramatically enhancing texture realism and fidelity but inducing shape distortions. Informed by this key insight, we introduce the Target-Balanced Score Distillation (TBSD). It formulates generation as a multi-objective optimization problem and introduces an adaptive strategy that effectively resolves the aforementioned trade-off. Extensive experiments demonstrate that TBSD significantly outperforms existing state-of-the-art methods, yielding 3D assets with high-fidelity textures and geometrically accurate shape.