A Universal Avoidance Method for Diverse Multi-branch Generation

TL;DR

UAG is a model-agnostic, efficient method that significantly improves diversity in generative models by penalizing output similarity, outperforming existing approaches in speed and computational cost.

Contribution

We propose UAG, a novel universal strategy that enhances multi-branch diversity in generative models with minimal additional computation and broad applicability.

Findings

Achieves up to 1.9x higher diversity

Runs 4.4x faster than state-of-the-art methods

Uses only 1/64 of the FLOPs compared to existing approaches

Abstract

Modern generative models still lack human-level creativity, particularly in multi-branch diversity. Prior approaches to address this problem often incur heavy computation or strong dependency on model architecture. Therefore, we introduce UAG(Universal Avoidance Generation), a model-agnostic and computationally efficient generation strategy that penalizes similarity among previously generated outputs. Thus, UAG can enhance multi-branch diversity across both diffusion and transformer models, with minimal additional computation. In experiments, our method achieves up to 1.9 times higher diversity, runs 4.4 times faster, and requires only 1/64 of the FLOPs compared to state-of-the-art methods. The full code is https://anonymous.4open.science/r/2026_ACL_Universal/.