FlowTranslator enables training-free in-image machine translation using flow-based editing

FlowTranslator achieves average PRIM BLEU and COMET scores of 20.6 and 57.2 on EasyOCR read-back protocol and 40.1% exact-match accuracy on ScenePair photographs translated into Chinese.

In-image machine translation replaces text in an image with its translation while preserving the surrounding visual context. FlowTranslator is a training-free IIMT framework that first employs an OCR model and an LLM to generate the translation prompt which is injected into the T2I-based diffusion model. Two trajectory controls optimize text generation: Region Coupling Release partially decorrelates target-branch noise inside the text region early in the trajectory while Coupling Expansion amplifies the emerging edit displacement mid-trajectory. Under the EasyOCR read-back protocol it achieves average PRIM BLEU and COMET scores of 20.6 and 57.2 respectively. On ScenePair photographs translated into Chinese it reaches 40.1% exact-match accuracy and surpasses existing editing systems.