AnchorDS: Anchoring Dynamic Sources for Semantically Consistent Text-to-3D Generation
This work solves the issue of semantic over-smoothing artifacts in text-to-3D generation for applications like 3D content creation, though it is incremental as it builds on existing optimization-based methods.
The paper tackled the problem of semantic inconsistency in text-to-3D generation by addressing the static treatment of guidance in Score Distillation Sampling, resulting in AnchorDS, which improves detail, color, and semantic consistency for complex prompts while maintaining efficiency.
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.