Geometric Disentanglement of Text Embeddings for Subject-Consistent Text-to-Image Generation using A Single Prompt

TL;DR

This paper introduces a training-free geometric method to refine text embeddings, significantly enhancing subject consistency and text alignment in text-to-image generation without additional training or fine-tuning.

Contribution

It proposes a novel geometric approach to disentangle text embeddings, addressing semantic leakage and improving subject consistency in a training-free manner.

Findings

Improves subject consistency across generated images.

Enhances text alignment in text-to-image models.

Outperforms existing baselines in experiments.

Abstract

Text-to-image diffusion models excel at generating high-quality images from natural language descriptions but often fail to preserve subject consistency across multiple outputs, limiting their use in visual storytelling. Existing approaches rely on model fine-tuning or image conditioning, which are computationally expensive and require per-subject optimization. 1Prompt1Story, a training-free approach, concatenates all scene descriptions into a single prompt and rescales token embeddings, but it suffers from semantic leakage, where embeddings across frames become entangled, causing text misalignment. In this paper, we propose a simple yet effective training-free approach that addresses semantic entanglement from a geometric perspective by refining text embeddings to suppress unwanted semantics. Extensive experiments prove that our approach significantly improves both subject consistency and text alignment over existing baselines.