OmniPresent turns research papers into slides, posters, and videos

Transforming a static research paper into a dynamic presentation suite—posters, slides, and videos—is a labor-intensive task that existing tools handle only in isolation, often breaking semantic consistency. A team of researchers from multiple institutions (lead author Qianli Ma) has introduced OmniPresent, a unified framework that generates coherent presentations from scientific papers. The approach uses a renderable HTML representation to centralize content planning, enabling consistent messaging across all output formats. A verify-and-repair loop actively detects and resolves conflicts between modalities, such as mismatched figures or inconsistent text, ensuring the final deliverables are faithful to the source paper.

To support further research, the team released OmniPreBench, a comprehensive dataset containing over 1,000 scientific papers with paired artifacts (posters, slides, videos). They also established a VLM-based evaluation protocol that assesses both accuracy and visual appeal. Empirical results show that OmniPresent significantly outperforms strong baselines, producing high-quality, consistent presentation suites. This work formalizes the task of unified presentation suite generation and provides tools for scalable research, making it easier for scientists to disseminate their work effectively across multiple channels.