WEAVE: Unleashing and Benchmarking the In-context Interleaved Comprehension and Generation

TL;DR

WEAVE introduces a comprehensive benchmark suite and large-scale dataset for multi-turn, context-dependent multimodal comprehension and generation, addressing a key gap in existing visual understanding research.

Contribution

The paper presents WEAVE, the first dataset and benchmark for in-context interleaved multimodal comprehension and generation tasks, enabling better evaluation of multi-turn, context-aware models.

Findings

Training on WEAVE-100k improves multimodal understanding and editing capabilities.

Models develop emergent visual-memory abilities with WEAVE training.

Current models still face significant challenges in multi-turn, context-aware image generation.

Abstract

Recent advances in unified multimodal models (UMMs) have enabled impressive progress in visual comprehension and generation. However, existing datasets and benchmarks focus primarily on single-turn interactions, failing to capture the multi-turn, context-dependent nature of real-world image creation and editing. To address this gap, we present WEAVE, the first suite for in-context interleaved cross-modality comprehension and generation. Our suite consists of two complementary parts. WEAVE-100k is a large-scale dataset of 100K interleaved samples spanning over 370K dialogue turns and 500K images, covering comprehension, editing, and generation tasks that require reasoning over historical context. WEAVEBench is a human-annotated benchmark with 100 tasks based on 480 images, featuring a hybrid VLM judger evaluation framework based on both the reference image and the combination of the original image with editing instructions that assesses models' abilities in multi-turn generation, visual memory, and world-knowledge reasoning across diverse domains. Experiments demonstrate that training on WEAVE-100k enables vision comprehension, image editing, and comprehension-generation collaboration capabilities. Furthermore, it facilitates UMMs to develop emergent visual-memory capabilities, while extensive evaluations on WEAVEBench expose the persistent limitations and challenges of current approaches in multi-turn, context-aware image generation and editing. We believe WEAVE provides a view and foundation for studying in-context interleaved comprehension and generation for multi-modal community.