Direction, not prompting
The models are good enough to recognize exact cinematography directions — lens, blocking, coverage, light. I direct them like crew and hold them to the brief.
Twenty years of production craft behind the lens and in the timeline. Today's models take exact cinematography direction — lens, blocking, light, coverage — so I direct them like crew, then finish by hand in DaVinci Resolve Studio and the Adobe Suite.
Greek electronic track, anime-photorealistic hybrid, warm summer Athens. The video's real weight was upstream: it played a major role in visualizing the desired outcome — a concrete, watchable reference for what the finished piece should become. Click to load.
The models are good enough to recognize exact cinematography directions — lens, blocking, coverage, light. I direct them like crew and hold them to the brief.
DaVinci Resolve Studio for conform and grade, Adobe Suite for design layers, ComfyUI and LTX-Video locally for generation, cloud models where scale earns it. Local-first, tool-agnostic.
Sprite sheets sliced and framed by script, mosaics grown from seeded Python, video finished by deterministic ffmpeg passes. Repeatable by design.
An autonomous operator agent runs the render queue, verifies outputs and reports — the boring 80% is automated so the hand-work stays hand-work.