Scene text detection, which is one of the most popular topics in both academia and industry, can achieve remarkable performance with sufficient training data. However, the annotation costs of scene text detection are huge with traditional labeling methods due to the various shapes of texts. Thus, it is practical and insightful to study simpler labeling methods without harming the detection performance. In this paper, we propose to annotate the texts by scribble lines instead of polygons for text detection. It is a general labeling method for texts with various shapes and requires low labeling costs. Furthermore, a weakly-supervised scene text detection framework is proposed to use the scribble lines for text detection. The experiments on several benchmarks show that the proposed method bridges the performance gap between the weakly labeling method and the original polygon-based labeling methods, with even better performance. We will release the weak annotations of the benchmarks in our experiments and hope it will benefit the field of scene text detection to achieve better performance with simpler annotations.