For successful scene text recognition (STR) models, synthetic text image\ngenerators have alleviated the lack of annotated text images from the real\nworld. Specifically, they generate multiple text images with diverse\nbackgrounds, font styles, and text shapes and enable STR models to learn visual\npatterns that might not be accessible from manually annotated data. In this\npaper, we introduce a new synthetic text image generator, SynthTIGER, by\nanalyzing techniques used for text image synthesis and integrating effective\nones under a single algorithm. Moreover, we propose two techniques that\nalleviate the long-tail problem in length and character distributions of\ntraining data. In our experiments, SynthTIGER achieves better STR performance\nthan the combination of synthetic datasets, MJSynth (MJ) and SynthText (ST).\nOur ablation study demonstrates the benefits of using sub-components of\nSynthTIGER and the guideline on generating synthetic text images for STR\nmodels. Our implementation is publicly available at\nhttps://github.com/clovaai/synthtiger.\n