Video scene text editing benchmark measures quality and stability trade offs
ViTeX-Bench: Benchmarking High-Fidelity Video Scene Text Editing
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Editing text in videos, like changing words on signs or labels, is tricky because the changes must look good and stay consistent as the video moves. The authors created ViTeX-Bench, a set of videos and tests to see how well different editing methods work for this task. They measured how accurate the text changes are, how smooth the edits look over time, and how well the rest of the scene stays untouched. They also released a new open-source video text editor that performs well in some key tests. ViTeX-Bench helps people build and compare better video text editing tools.
What this means in practice
- •For video content producers: Assess and improve tools that change printed or displayed text in videos while keeping natural motion and scene appearance.
- •For digital marketing teams: Use reliable video text editing to customize product labels or advertisements in videos with consistent and high-quality results.$Commercial implications: Enables creation and sale of editing services or software that modify text in commercial videos while preserving realism.
Authors
Xinghao Chen, Xiangbo Gao, Jiongze Yu, Yuheng Wu, Zhengzhong Tu
Abstract
Recent video generation is increasingly realistic and controllable, yet video editing remains less developed, particularly for precise local edits that must preserve the original scene dynamics. Video scene text editing replaces text on scene surfaces, such as storefront signs, whiteboards, and product labels, while preserving the surrounding content, motion, and camera dynamics. Although scene text editing is well studied for images, video scene text editing that achieves high visual quality, temporal consistency, and edit locality remains underexplored. Existing resources offer limited paired real-video data, and general video-editing metrics do not directly measure whether the requested text remains correct over time. We introduce ViTeX-Bench, a benchmark suite comprising ViTeX-Dataset and a three-axis evaluation protocol. The dataset contains 387 real-world 720p videos with text-region masks and editing instructions: 230 provide reviewed, pipeline-generated paired edits for training, and 157 form a frozen evaluation split. The protocol evaluates text correctness, visual and temporal quality, and edit locality through 13 metrics, with one primary metric per axis and a Pareto comparison of their trade-offs. OCR calibration, human evaluation, and annotation-sensitivity analyses support the interpretation of these scores. Across eight baselines from four editing families, accurate text, temporal stability, and scene preservation remain difficult to achieve together. We also release ViTeX-Edit-14B, an open-source reference editor fine-tuned on the paired training split with motion-aligned glyph-video conditioning. It achieves CharAcc 0.688, the highest mean among the evaluated video-native editors, and the lowest comparable text-crop Warp among raw editor outputs. ViTeX-Bench provides a reproducible foundation for studying these trade-offs in video scene text editing.