Write HTML. Render video. Built for Agents.
The Hook
“Motion design isn't dead. In fact, we think it's just starting to expand and AI will play a huge part in unlocking the possibility for more people to create video.”
Problem Agitation
“You see, every new model comes with a report card, a table of scores against benchmarks. These tests measure how well a model solves hard, checkable problems, like code and agentic work or math and science.”
Problem Agitation
“But many people are pushing these models somewhere else that is much harder to quantify. Creative work, UI design, writing, video. There's no answer key for those. So people improve by trial and error. Our team set out to change that for video generated from code, a quantifiable way to measure how well a model makes video.”
Product Reveal
“Introducing the Code2Video benchmark. I'm sure you're asking, well, who says a video is good?”
Feature Teaser
“We started with people, two generated videos side by side followed by simple questions. Does it hold your attention? Does it match the brief? Thousands of judgments, all made by humans. Then we trained a model on those picks. Show it two videos and a question, and it tells you which one people would choose and why. This model flickers. That one ignores the brief. And because the judge can score any two videos in milliseconds, models can train against it, not just be ranked by it.”
Call to Action
“Over time, this benchmark will help models produce better videos from code. This is just the beginning. Hyperframes, go make something.”
Synthesizable in Remotion · key components of the code2video-benchmark-reveal template.
A central 3D wireframe sphere composed of dot particles rotating and pulsing to represent the AI judge node.
Render SVG circle elements or a Three.js PointCloud in Remotion; map coordinates with spherical trigonometry and drive rotation via currentFrame.
Two split UI canvas windows presented side-by-side with synchronized animated playback and selection buttons.
Wrap dual video/canvas layers in flex containers, applying spring animations on mount and highlight borders based on interactive cursor triggers.
Rapid vertical terminal text scroll displaying evaluation logs, model comparisons, and selection outcomes.
Translate a multi-row mono-spaced data table vertically using interpolate(frame, [start, end], [0, -totalHeight]) with an active row highlight.
Purpose. Capture motion designers and developers by challenging the 'AI kills motion' meme.
Execution. Rapidly stack authentic tweet screenshots over dark background with motion reels.
Purpose. Explain why AI excels at code/math but struggles with creative and motion domains.
Execution. Contrast formal benchmark tables with an unquantifiable mosaic of creative web and UI clips.
Purpose. Present Code2Video's methodology: human preference dataset converted into an AI reward model.
Execution. Display side-by-side video A/B evaluations linked to a glowing point-cloud judge.
Purpose. Establish authority via DeepMind partnership and call community to build with the tool.
Execution. Interactive bezier editor graphic followed by dual logo co-branding and Kaggle availability tag.