Diverse Text-to-Image Generation via Contrastive Noise Optimization

Researchers introduce Contrastive Noise Optimization, a new method that improves diversity in text-to-image AI generation by optimizing initial noise patterns rather than intermediate outputs. The technique uses contrastive loss to maximize diversity while preserving image quality, achieving superior results across multiple text-to-image model architectures.

• →Contrastive Noise Optimization addresses the limited diversity problem in text-to-image diffusion models by shaping initial noise rather than intermediate latents.
• →The method uses contrastive loss in Tweedie data space to repel instances within a batch while maintaining fidelity to reference samples.
• →Extensive experiments show the approach achieves superior quality-diversity balance across multiple T2I model backbones.
• →The technique is robust to hyperparameter choices, making it more practical than existing diversity enhancement methods.
• →The research provides theoretical insights explaining why preprocessing noise is effective for improving output diversity.