8 upvotes
IntReal introduces real-world benchmark for multimodal interleaved generation
What is IntReal: A Real-World Dataset and Benchmark for Multimodal Interleaved Generation?
The paper releases IntReal, a dataset of text and images that are naturally mixed, plus a set of evaluation tasks for models that generate both modalities in sequence. It lets researchers test how well their unified multimodal models handle the back-and-forth flow of real content.
Who it is for
Anyone building or evaluating models that need to output text and pictures together should use it to gauge real-world performance.