Search papers, labs, and topics across Lattice.
This paper introduces T2LSC-Bench, a benchmark designed to evaluate localized semantic control in text-to-image generation, addressing the issue of target-text-associated semantic leakage. The study reveals that while text rendering accuracy remains high, semantic leakage can significantly increase under stress-test conditions, indicating that accurate text generation does not ensure the preservation of surrounding scene semantics. Notably, the implementation of anti-leakage prompting effectively reduces semantic leakage rates without compromising rendering quality, highlighting a critical gap in current evaluation methods for text-to-image models.
Accurate text rendering in image generation doesn't prevent significant semantic leakage, with rates spiking from 1.2% to 18.1% under stress conditions.
Recent text-to-image models have become increasingly capable of rendering explicit text, but reliable localized text control requires more than generating the correct string. In applications such as product labeling, signage, and interface design, target text should be rendered within a designated text-bearing region without altering the predefined subject identity or surrounding scene semantics. We refer to violations of this requirement as target-text-associated semantic leakage, in which target-text semantics are expressed through non-textual visual content beyond the designated anchor. Existing visual-text benchmarks primarily evaluate readability, spelling accuracy, and layout, leaving this form of semantic leakage largely unexamined. We introduce T2LSC-Bench, a controlled diagnostic benchmark comprising 50 seed subjects and 1,200 prompt cases per model, yielding 7,160 evaluated images across six models. Its factorized design varies semantic relation, scene openness, prompt mode, and language. A dual-branch protocol combines OCR-VLM text verification with structured VLM semantic judgments to measure Text-at-Anchor Accuracy (TAA), Semantic Subject Preservation (SSP), Semantic Leakage Rate (SLR), and Conditional Semantic Leakage Rate (cSLR). Under stress-test conditions, SLR increases from 1.2% to 18.1% and cSLR from 1.3% to 18.2%, whereas TAA decreases only from 91.4% to 90.9%. Anti-leakage prompting reduces SLR from 16.6% to 8.4% without degrading rendering accuracy. Human validation on 420 images shows strong agreement between automatic and adjudicated annotations. These results show that accurate text rendering does not guarantee local containment of target-text semantics.
