Researchers have introduced MotionGraphicsBench, a new benchmark designed to evaluate the performance of large language models in generating motion graphics from static images and natural language prompts. This benchmark consists of 100 curated instances, each paired with a programmatic verification system that assesses adherence to spatial, appearance, and temporal requirements. The verification programs, developed using a multi-agent pipeline, achieve high agreement with human judgments and outperform existing vision-language model judges. The benchmark was used to test 32 LLMs across 15 models, revealing a wide range in generation capabilities, and was also employed to fine-tune the Qwen3.8-27B model, significantly improving its performance. AI
IMPACT This benchmark could drive improvements in AI-generated animation and design tools by providing a standardized evaluation method.
RANK_REASON The cluster describes a new academic benchmark for evaluating AI model capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →