Elucera

Elucera turns a single topic into a finished 1080×1920 science or computer-science explainer Reel with captions. There is no AI-generated video: every frame is rendered from code. You type a topic and Muse Spark writes a four-part script (Hook, Setup, Mechanism, Payoff) and pitches a storyboard. You can edit the narration or add a director’s note, then approve. Next, the pipeline records a voiceover and gets word-level timings with faster-whisper. A scene planner then chooses an animation engine for each beat: Manim for math and mechanism diagrams, Remotion for kinetic type and charts, Three.js for 3D, or a p5.js doodle renderer. Muse writes the code for each scene, and the scenes render in parallel inside sandboxes (Manim runs in Docker with no network). If a render fails, the error goes back to the model and it fixes its own code, up to 3 times. FFmpeg then puts the clips together and burns in word-highlight captions timed to the voice. Each successful scene and each fixed error is saved to a library that feeds future prompts, so the system improves with every run. You can redo a single scene with a note, compare before and after, and rebuild the reel without regenerating the rest. A live run goes from topic to finished reel in about 4 minutes.