With the development of Artificial Intelligence Generated Content (AIGC) technologies, video generation quality assessment has become an emerging research topic. Traditional video quality assessment methods often rely on manual evaluation or single-modal automated metrics, which suffer from inefficiency, subjectivity, and the inability to comprehensively assess video content. To address these issues, this paper proposes a multi-modal quality assessment benchmark based on semantic distance of video frames. The method first computes the semantic distance of each frame based on its semantic features and then partitions the video into several “clusters” with semantically similar frames. Within each group, representative frames are selected for frame extraction, and similarity scores between frames are computed. Based on frame extraction and similarity calculation, existing image quality assessment benchmarks are further applied to evaluate the quantity, subject, and color of video frames. Experimental results demonstrate that the proposed method significantly improves the accuracy and consistency of video quality assessment, especially in terms of semantic consistency and the diversity of visual content, outperforming traditional single-modal assessment methods. This work provides a new automated evaluation approach for video generation quality, with strong practical value and research potential.