SlideLab: Audience-Centered Scientific Slide Generation Framework

What happened

An R&D team has launched SlideLab, a training-free multi-agent framework that generates audience-tailored academic presentation decks directly from research papers. The framework first plans the presentation narrative, then iteratively optimizes a shared slide deck across four specialized agents: content planner, visual generator, layout refiner, and implementation validator. Blind testing shows the system outperforms open-source and commercial systems on 77% of paper generation tasks, uses roughly 4x fewer inference tokens than the strongest open-source baseline, and launches ConfArena, a companion audience-focused presentation evaluation framework alongside it.

Key facts

Framework type

Training-free multi-agent academic presentation generation framework

Blind test preference rate

Outperforms open-source and commercial systems on 77% of paper generation tasks

Inference token consumption

Roughly 4x lower than the strongest open-source baseline

Companion evaluation framework

ConfArena, which can simulate a conference room setting to evaluate presentations page by page

Evaluation framework capabilities

Can detect presentation issues including fabricated figures, missing pages, and out-of-order content

Background

Academic presentations are not simple summaries of research papers; they require building a coherent narrative and clearly explaining core arguments to help audiences follow along smoothly. Generating and evaluating such presentations has long posed a high professional barrier for researchers.

Why it matters

For the research community, the training-free framework drastically reduces the time cost for researchers to build high-quality presentation decks, while the companion ConfArena evaluation framework provides measurement metrics aligned with real conference scenarios. For developers, the training-free multi-agent architecture requires no additional training investment, can be quickly integrated into existing research toolchains, and lowers the barrier to secondary development.