Building the tools, not just using them.
Fifteen years in Cinema 4D and After Effects, and a generation of AI models that do not talk to either. So I started building the connections myself.
LK Motion Viewport AI is a Cinema 4D plugin in development. Exporting a playblast to Runway or a similar tool is easy and it works, but the result lands outside the scene with no per-object control. This runs inside Cinema 4D instead. It is a working experiment, not a product, and none of it is for sale.
In development
Viewport in, finished frame out
The raw viewport render becomes a finished image in the same panel, with a wipe to compare the two side by side.
Per-object control by colour
Objects are assigned or painted a key colour, and each colour carries its own prompt. Red becomes the jellyfish, green becomes the lights. The pass works to exact masks rather than guessing at the scene.
Local or cloud, structure held
Offline local models for iteration, or a cloud engine for final quality. Geometry and seed both lock, so the look changes while the scene and the result stay repeatable.
Output returns to the scene
Materials, 360 skies and editable Octane scenes from the same prompt. Materials already generate on the offline engine, so the output comes back as a usable asset rather than stopping at a picture.
Not built yet: live viewport preview through fal.ai, deeper Seedance integration, and a wider set of offline engines.
Why this matters
Using AI tools is common now. Building the layer underneath them is a different problem: exact masks per object, model routing, structure and seed locking, local and cloud engines, and getting a result back into an editable scene. Working at that level is what makes generative AI useful on client work rather than incidental to it. The plugin is built solo, with AI-assisted development.