AI-generated stillHow a single still becomes a moving shot: what the model does, how to direct the camera with words, and what usually goes wrong.
In Cinetana's Studio, a shot starts as a still picture made with Google's Gemini image model. The best stills are then sent to Kling, a third-party video model, with a short motion prompt: push in slowly, let the waves break, let the smoke drift.
It works best on scenes with one clear subject and a natural motion. It still fails in familiar ways: hands that change shape, faces that drift, text that melts. This topic shows those faults on our own frames, and the fixes we use.