AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation

TL;DR

AnchorDS introduces a novel approach to text-to-3D generation that dynamically anchors source distributions to improve semantic consistency, detail, and color accuracy in generated 3D models, especially for complex prompts.

Contribution

The paper proposes AnchorDS, a new score distillation method that dynamically anchors source distributions using image conditions, enhancing semantic consistency and detail in text-to-3D generation.

Findings

Produces finer-grained detail and more natural colors.

Achieves stronger semantic consistency for complex prompts.

Surpasses previous methods in quality and efficiency.

Abstract

Optimization-based text-to-3D methods distill guidance from 2D generative models via Score Distillation Sampling (SDS), but implicitly treat this guidance as static. This work shows that ignoring source dynamics yields inconsistent trajectories that suppress or merge semantic cues, leading to "semantic over-smoothing" artifacts. As such, we reformulate text-to-3D optimization as mapping a dynamically evolving source distribution to a fixed target distribution. We cast the problem into a dual-conditioned latent space, conditioned on both the text prompt and the intermediately rendered image. Given this joint setup, we observe that the image condition naturally anchors the current source distribution. Building on this insight, we introduce AnchorDS, an improved score distillation mechanism that provides state-anchored guidance with image conditions and stabilizes generation. We further penalize erroneous source estimates and design a lightweight filter strategy and fine-tuning strategy that refines the anchor with negligible overhead. AnchorDS produces finer-grained detail, more natural colours, and stronger semantic consistency, particularly for complex prompts, while maintaining efficiency. Extensive experiments show that our method surpasses previous methods in both quality and efficiency.