Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation

TL;DR

This paper extends one-step image generation from class labels to text inputs by integrating LLM-based text encoders into the MeanFlow framework, enabling richer content creation with improved performance.

Contribution

It introduces a novel approach to incorporate powerful text encoders into MeanFlow for effective text-conditioned image synthesis, addressing challenges of discriminability in limited-step generation.

Findings

Successful adaptation of MeanFlow for text-conditioned image generation.

Significant improvements in generation quality demonstrated on diffusion models.

Analysis reveals the importance of high discriminability in text features for limited-step generation.

Abstract

Few-step generation has been a long-standing goal, with recent one-step generation methods exemplified by MeanFlow achieving remarkable results. Existing research on MeanFlow primarily focuses on class-to-image generation. However, an intuitive yet unexplored direction is to extend the condition from fixed class labels to flexible text inputs, enabling richer content creation. Compared to the limited class labels, text conditions pose greater challenges to the model's understanding capability, necessitating the effective integration of powerful text encoders into the MeanFlow framework. Surprisingly, although incorporating text conditions appears straightforward, we find that integrating powerful LLM-based text encoders using conventional training strategies results in unsatisfactory performance. To uncover the underlying cause, we conduct detailed analyses and reveal that, due to the extremely limited number of refinement steps in the MeanFlow generation, such as only one step, the text feature representations are required to possess sufficiently high discriminability. This also explains why discrete and easily distinguishable class features perform well within the MeanFlow framework. Guided by these insights, we leverage a powerful LLM-based text encoder validated to possess the required semantic properties and adapt the MeanFlow generation process to this framework, resulting in efficient text-conditioned synthesis for the first time. Furthermore, we validate our approach on the widely used diffusion model, demonstrating significant generation performance improvements. We hope this work provides a general and practical reference for future research on text-conditioned MeanFlow generation. The code is available at https://github.com/AMAP-ML/EMF.