DCText: Scheduled Attention Masking for Visual Text Generation via Divide-and-Conquer Strategy

TL;DR

DCText introduces a divide-and-conquer approach with scheduled attention masking and localized noise initialization to improve long and multi-text visual generation, achieving high accuracy and low latency.

Contribution

It presents a novel training-free method that decomposes prompts and applies attention masks for better text rendering in images, outperforming existing models.

Findings

Achieves superior text accuracy on benchmarks.

Maintains high image quality with lower latency.

Effective for both single- and multi-sentence prompts.

Abstract

Despite recent text-to-image models achieving highfidelity text rendering, they still struggle with long or multiple texts due to diluted global attention. We propose DCText, a training-free visual text generation method that adopts a divide-and-conquer strategy, leveraging the reliable short-text generation of Multi-Modal Diffusion Transformers. Our method first decomposes a prompt by extracting and dividing the target text, then assigns each to a designated region. To accurately render each segment within their regions while preserving overall image coherence, we introduce two attention masks - Text-Focus and Context-Expansion - applied sequentially during denoising. Additionally, Localized Noise Initialization further improves text accuracy and region alignment without increasing computational cost. Extensive experiments on single- and multisentence benchmarks show that DCText achieves the best text accuracy without compromising image quality while also delivering the lowest generation latency.