inkling.

Problem solved

Most AI explainer tools hand back images that drift between frames, with labels the model garbles, and neither is editable afterwards. inkling keeps every scene as text, so the same input renders the same frames every time and you can review a change as a diff.

What I built

Hand-drawn animated explainer videos rendered from code, with no AI image model and no AI video model. Each scene is one animated SVG in a plain HTML file: CSS keyframes plus a recurring mascot. Puppeteer loads the scene, freezes the CSS clock frame by frame and screenshots a deterministic sequence, then ffmpeg stitches the scenes with crossfades into an mp4. Nine themes and eight mascot shapes are called by name, so the same scenes render in a different look by changing one attribute. A second path starts from a YouTube link: yt.js pulls the audio, transcribes it with local Whisper and extracts per-frame amplitude, your coding agent writes the scenes from the transcript, and the audio level is injected into every frame as a CSS variable the scenes animate against.