StreamEval: Evaluating Interactive Streaming Video Generation Models

One-line takeaway

StreamEval is a unified benchmark for interactive streaming video generation: 5 tasks, 300 scripted 15–300 s cases and 10 metrics; across 37 models, high visual quality and identity retention do not guarantee reliable responses to mid-stream requests.

StreamEval 是交互式流式视频生成的统一基准：5 类任务、300 个 15–300 秒脚本用例、10 项 0–100 指标，评测 37 个模型后发现画质高、身份保持好并不等于能可靠响应中途到来的指令。