Toward Guidance-Free AR Visual Generation via Condition Contrastive Alignment

TL;DR

This paper introduces Condition Contrastive Alignment (CCA), a guidance-free method for autoregressive visual generation that fine-tunes pretrained models to match target distributions, reducing reliance on guidance techniques and improving efficiency.

Contribution

CCA provides a guidance-free approach for AR visual generation, unifying language and visual alignment methods, and significantly reduces sampling costs with minimal fine-tuning.

Findings

CCA enhances guidance-free performance with just one epoch of fine-tuning.

CCA achieves results comparable to guided sampling methods.

Adjustable training parameters allow trade-offs between diversity and fidelity.

Abstract

Classifier-Free Guidance (CFG) is a critical technique for enhancing the sample quality of visual generative models. However, in autoregressive (AR) multi-modal generation, CFG introduces design inconsistencies between language and visual content, contradicting the design philosophy of unifying different modalities for visual AR. Motivated by language model alignment methods, we propose \textit{Condition Contrastive Alignment} (CCA) to facilitate guidance-free AR visual generation with high performance and analyze its theoretical connection with guided sampling methods. Unlike guidance methods that alter the sampling process to achieve the ideal sampling distribution, CCA directly fine-tunes pretrained models to fit the same distribution target. Experimental results show that CCA can significantly enhance the guidance-free performance of all tested models with just one epoch of fine-tuning ($\sim$ 1\% of pretraining epochs) on the pretraining dataset, on par with guided sampling methods. This largely removes the need for guided sampling in AR visual generation and cuts the sampling cost by half. Moreover, by adjusting training parameters, CCA can achieve trade-offs between sample diversity and fidelity similar to CFG. This experimentally confirms the strong theoretical connection between language-targeted alignment and visual-targeted guidance methods, unifying two previously independent research fields. Code and model weights: https://github.com/thu-ml/CCA.