Training-Free Generation of Diverse and High-Fidelity Images via Prompt Semantic Space Optimization
This addresses a fundamental challenge for creative exploration and downstream applications in AI image generation, though it is an incremental improvement over existing methods.
The paper tackles the problem of low diversity in text-to-image diffusion models, which generate repetitive outputs, by proposing TPSO, a training-free module that improves generative diversity from 1.10 to 4.18 points on benchmarks without sacrificing image quality.
Image diversity remains a fundamental challenge for text-to-image diffusion models. Low-diversity models tend to generate repetitive outputs, increasing sampling redundancy and hindering both creative exploration and downstream applications. A primary cause is that generation often collapses toward a strong mode in the learned distribution. Existing attempts to improve diversity, such as noise resampling, prompt rewriting, or steering-based guidance, often still collapse to dominant modes or introduce distortions that degrade image quality. In light of this, we propose Token-Prompt embedding Space Optimization (TPSO), a training-free and model-agnostic module. TPSO introduces learnable parameters to explore underrepresented regions of the token embedding space, reducing the tendency of the model to repeatedly generate samples from strong modes of the learned distribution. At the same time, the prompt-level space provides a global semantic constraint that regulates distribution shifts, preventing quality degradation while maintaining high fidelity. Extensive experiments on MS-COCO and three diffusion backbones show that TPSO significantly enhances generative diversity, improving baseline performance from 1.10 to 4.18 points, without sacrificing image quality. Code will be released upon acceptance.