Improving text-to-image generation by handling negation clearly
Negation without Confusion: Decomposing Bias and Recomposing Style in Text Embeddings for Negation-aware Image Generation
Computer Vision and Pattern Recognition
Summary
Text-to-image AI models often have difficulty understanding and correctly processing negation, like when a description says "no cat." This is because the way these models represent text mixes the actual meaning with biases that cause errors. The authors propose a method called Dero that separates meaning from bias in text descriptions and then recombines them to better respect negation without losing style. This method works without retraining and can be added to different existing models, leading to clearer, more accurate images that match the text better.
What this means in practice
- •For visual content creators: Generate images from text prompts that accurately reflect negations such as "no glasses" or "without hats" improving creative control.
- •For natural language processing engineers: Incorporate negation-aware text embedding refinement into image generation pipelines to improve semantic accuracy without retraining models.
Authors
Sooyeon Park, Hemraj Singh, Kyunghyun Min, Youngjae Yu, Sung-Bae Cho
Abstract
Negation-aware image generation remains challenging for text-to-image diffusion models, which often struggle with negation due to implicit linguistic biases that are deeply entangled within text representations. As a result, bias-aligned semantics frequently dominate over intended negative constraints. To overcome this limitation, we propose a simple training-free method of Decompose and Recompose text embedding (Dero), which constructs a semantic-preserving subspace and a bias subspace using auxiliary prompts and performs semantic disentanglement based on bidirectional SVD. By decomposing entangled representations and selectively recomposing disentangled semantic components, Dero realigns text representations toward intended negation semantics while restoring residual stylistic features to preserve semantic fidelity and image quality. As a model-agnostic inference-time framework operating solely on text representations, Dero can be seamlessly applied to diverse diffusion architectures, including both U-Net and DiT based models. Extensive experiments on both object- and concept-level negation demonstrate consistent improvements in negation-aware image generation, achieving stronger text-image alignment, better semantic fidelity, and higher human preference scores.