An agent for label-heavy technical animation.
Tandava is a 4-stage agentic pipeline — concept → image → accuracy gate → motion — that turns Runway's flagship video model into a reliable instrument. The same loop ships across science, engineering, and performance art with text-stable labels, biophysically correct motion, and zero hallucinated structures.
Tandava makes label-heavy technical diagrams move without deforming the labels, hallucinating new structures, or breaking the biophysics. Three domains, one loop, end-to-end agentic.
The 4-stage pipeline
Every clip on this page traverses the same four stages. The accuracy gate iterates back into the same OpenAI conversation via the captured URL, sending diff-only structural refinements — never re-prompting the full body.
Domain 1 — Science 11 clips · 35 production renders on veo3.1 @ 1920×1080
Eleven label-dense scientific cross-sections spanning molecular biology, virology, geophysics, astrophysics, theoretical relativity, planetary science, atmospheric dynamics, and biophysics. Text remains pixel-stable, biophysics stays correct, no new structures appear. Cosmological and atmospheric clips use the upscale → animate workflow: user-edited 4:3 source → Real-ESRGAN x4 → §14 motion scaffold with physically calibrated rates (3–4 Hz flame flicker, prograde differential disk rotation, traveling-wave envelope on basilar membrane, frame-dragging spiral, mesocyclone rotation, subduction-zone drift).
Domain 2 — Engineering 3 systems · vertical farm + pipeline schematic + smart-ring
Three engineering cross-sections that exercise the same agentic pipeline applied to totally different scales and modalities — a building-scale recirculating ecosystem, the studio's own data-flow architecture, and a wearable health monitor down at the millimetre scale. The smart-ring was refined v1→v2 in the same OpenAI conversation: diff-only structural updates (photodiode redrawn as receiver, 940nm IR LED made invisible, 1mm scale bar added) — zero re-prompt.
Domain 3 — Performance Art Indian classical dance & yoga anatomy · 4 poses · veo3.1 @ 1920×1080
Four label-dense pose-anatomy plates from the Bharatanatyam and yoga lineages. The same §14 scaffold that locks Earth's magnetosphere in place also locks a tribhanga stance, a baddhakonasana opening, an alapadma mudra, and a swastikasana seat — generality proven across radically different subject matter. Sanskrit labels with English transliteration · muscle groups, energy lines, and weight-transfer axes frozen · subtle breath cycle is the only motion.
The §14 hardened scaffold
Naive prompts on dense technical diagrams fail in three predictable ways: text glyphs deform, the camera ignores "static" hints and zooms anyway, and subjects animate poetically instead of physically. Tandava enforces a hardened scaffold on every submission, distilled across 35 production renders.
{key}_v{VERSION}.mp4 · never overwrite a prior take