HyperFrames

HyperFrames is an open-source framework from HeyGen that lets AI coding agents generate video by writing HTML, CSS, and JavaScript, which then render deterministically into MP4 files. It is Apache-2.0 licensed, built to be authored by AI coding agents such as Claude Code and Codex , and aimed at developers building agentic video pipelines rather than teams clicking through timeline editors. HyperFrames pioneered HTML-to-video and has climbed past 50k GitHub stars in just a few months , which sets the context for this launch. The launch itself introduces the Code2Video Benchmark, released in collaboration with Google DeepMind and Kaggle. The task is straightforward: given a prompt or visual narrative, an LLM agent generates code that a rendering framework such as HyperFrames deterministically renders into the final video. The problem the team is trying to solve is that while modern LLMs are proficient at writing valid code, they struggle with motion graphics and visual intent, and writing code that executes is fundamentally different from crafting motion design that looks compelling, balanced, and professionally animated . To score that, the team built a pairwise judge trained on thousands of side-by-side human comparisons, so any two generated videos can be ranked in milliseconds and models can train against the signal rather than only be graded by it. For founders and researchers working on agentic creative tools, the launch matters because it turns a previously vibes-based capability into something measurable, with a public leaderboard and dataset. HeyGen partnered with Kaggle to release the full Code2Video Benchmark, including all reference compositions, design briefs, and the evaluation judge API, with every model run in the report executed on the Kaggle platform . The research is credited to Abhay Zala, Miao Yang, Wenbo Zhu, and Bin Liu alongside the Kaggle team, extending HeyGen's earlier open-source push around HyperFrames itself.