VDE Bench: Evaluating The Capability of Image Editing Models to Modify Visual Documents
Researchers have introduced new benchmarks and datasets for evaluating image editing models, addressing limitations in current systems. VINS-120K offers a large-scale dataset for ultra-high-resolution image editing, while VDE Bench focuses on modifying visual documents with dense text in multiple languages. VIBE, another benchmark, assesses models' ability to follow visual instructions, revealing that proprietary models currently outperform open-source alternatives but still struggle with complex tasks. Additionally, Together AI has launched FLUX.1 Kontext models, which enable in-context image generation and editing using both text and image prompts without requiring fine-tuning. AI
IMPACT New benchmarks and models are pushing the boundaries of AI image editing, enabling more precise control and higher resolutions.