TextCrafter: Accurately Rendering Multiple Texts in Complex Visual Scenes

TL;DR

TextCrafter tackles Complex Visual Text Generation by decomposing multi-text rendering into Instance Fusion, Region Insulation, and Text Focus within a training-free diffusion framework. It binds text content to visual carriers, isolates regional prompts, and strengthens attention on text regions, aided by a MILP-based layout optimizer and a new CVTG-2K benchmark. Empirical results on CVTG-2K and MARIOEval show superior text fidelity and aesthetic quality compared with state-of-the-art baselines, with strong ablations supporting component effectiveness. The work advances practical visual-text rendering in complex scenes and provides a scalable benchmark for future CVTG research.

Abstract

This paper explores the task of Complex Visual Text Generation (CVTG), which centers on generating intricate textual content distributed across diverse regions within visual images. In CVTG, image generation models often rendering distorted and blurred visual text or missing some visual text. To tackle these challenges, we propose TextCrafter, a novel multi-visual text rendering method. TextCrafter employs a progressive strategy to decompose complex visual text into distinct components while ensuring robust alignment between textual content and its visual carrier. Additionally, it incorporates a token focus enhancement mechanism to amplify the prominence of visual text during the generation process. TextCrafter effectively addresses key challenges in CVTG tasks, such as text confusion, omissions, and blurriness. Moreover, we present a new benchmark dataset, CVTG-2K, tailored to rigorously evaluate the performance of generative models on CVTG tasks. Extensive experiments demonstrate that our method surpasses state-of-the-art approaches.

Figures (7)

• Figure 1: TextCrafter enables precise multi-region visual text rendering, addressing the challenges of long, small-size, various numbers, symbols and styles in visual text generation. We illustrate the comparisons among TextCrafter with three state-of-the-art models, i.e., FLUX flux, TextDiffuser-2 chen2025textdiffuser and 3DIS zhou20243dis.
• Figure 2: Overview of our TextCrafter. TextCrafter consists of three steps. (a) Instance Fusion: strengthen the connection between visual text and its corresponding carrier. (b) Region Insulation: leverage the positional priors of the pre-trained DiT model to initialize the layout information for each text instance while separating and denoising text prompts across different regions. (c) Text Focus: enhance the attention maps of visual text, refining the fidelity of text rendering.
• Figure 3: Illustration of tokenizing the prompt “A sidewalk poster with ‘Register Now for arXiv’.” along with the attention map corresponding to each token. The use of preceding quotation marks can reinforce the relationship between text tokens and carrier tokens.
• Figure 4: Motivation for layout optimizer. LLM/MLLM-based layout designers frequently generate overlapping bounding boxes, which do not align with the layout preferences of Diffusion models. Our proposed layout optimizer addresses this limitation without additional fine-tuning.
• Figure 5: For a pre-trained DiT model, only a few denoising steps are required to approximate the image layout and subject positions. After 8 steps, the layout closely resembles that of a full 50-step process, with subsequent iterations primarily refining details.