OmniCaptioner: One Captioner to Rule Them All

We propose OmniCaptioner, a versatile visual captioning framework for generating fine-grained textual descriptions across a wide variety of visual domains. Unlike prior methods limited to specific image types (e.g., natural images or geometric visuals), our framework provides a unified solution for captioning natural images, visual text (e.g., posters, UIs, textbooks), and structured visuals (e.g., documents, tables, charts). By converting low-level pixel information into semantically rich textual representations, our framework bridges the gap between visual and textual modalities. Our unified and powerful pretraining on diverse visual domains enables OmniCaptioner to transfer effectively across multiple downstream tasks. Specifically, it demonstrates strong performance in: (i) Enhanced Visual Reasoning with LLMs, where long-context captions of visual modalities empower LLMs, particularly the DeepSeek-R1 series, to reason effectively in multimodal scenarios; (ii) Improved Image Generation, where detailed captions improve tasks like text-to-image generation and image transformation; and (iii) Efficient Supervised Fine-Tuning (SFT), which enables faster convergence with less data. Extensive evaluations show that our generated captions not only outperform existing methods on multiple benchmarks but also receive favorable feedback in user studies, validating both the accuracy and usefulness of our captions in real-world applications. We believe the versatility and adaptability of OmniCaptioner can offer a new perspective for bridging the gap between language and visual modalities.

arXiv preprint