The paper introduces *ChronoMagic-Bench*, a novel benchmark for evaluating text-to-video (T2V) models in generating time-lapse videos. Unlike existing benchmarks that focus on visual quality and textual relevance, *ChronoMagic-Bench* emphasizes the models' ability to produce videos with significant metamorphic amplitude and temporal coherence. The benchmark includes 1,649 prompts and real-world videos, categorized into four major types of time-lapse videos: biological, human-created, meteorological, and physical phenomena, with 75 subcategories. To assess the models' performance, two new automatic metrics, MTScore and CHScore, are introduced to evaluate metamorphic attributes and temporal coherence, respectively. The paper also presents *ChronoMagic-Pro*, a large-scale dataset containing 460k high-quality 720p time-lapse videos and detailed captions, designed to provide a comprehensive evaluation framework for T2V models. Comprehensive evaluations of ten representative T2V models using *ChronoMagic-Bench* reveal their strengths and weaknesses, highlighting the need for models to generate videos with rich physical content and temporal coherence. The paper concludes with a discussion of the limitations and future work, emphasizing the importance of addressing the challenges in generating metamorphic videos.The paper introduces *ChronoMagic-Bench*, a novel benchmark for evaluating text-to-video (T2V) models in generating time-lapse videos. Unlike existing benchmarks that focus on visual quality and textual relevance, *ChronoMagic-Bench* emphasizes the models' ability to produce videos with significant metamorphic amplitude and temporal coherence. The benchmark includes 1,649 prompts and real-world videos, categorized into four major types of time-lapse videos: biological, human-created, meteorological, and physical phenomena, with 75 subcategories. To assess the models' performance, two new automatic metrics, MTScore and CHScore, are introduced to evaluate metamorphic attributes and temporal coherence, respectively. The paper also presents *ChronoMagic-Pro*, a large-scale dataset containing 460k high-quality 720p time-lapse videos and detailed captions, designed to provide a comprehensive evaluation framework for T2V models. Comprehensive evaluations of ten representative T2V models using *ChronoMagic-Bench* reveal their strengths and weaknesses, highlighting the need for models to generate videos with rich physical content and temporal coherence. The paper concludes with a discussion of the limitations and future work, emphasizing the importance of addressing the challenges in generating metamorphic videos.