Vibe directing: the new posture

The whole point of the stack was leading here. When the tools take over enough of the pipeline, making a film stops being an act of operation and becomes an act of direction. This part is about that turn, the products driving it, and the argument it started.

The clearest sign that generation became a stack rather than a tool is a phrase that surfaced in mid 2026: vibe directingVibe directingDirecting a generative piece by intent and reference rather than by hand, describing the feel you want and steering it through iterations.. It is a deliberate echo of vibe coding, the idea from a year earlier that you can build software by describing what you want and letting the model write the code while you forget the code is even there. Vibe directing applies the same posture to film. OpenArt, the company that coined it, describes directing a film into existence by talking: you describe the movie, the system builds it across the whole pipeline, and you shape it conversationally, note by note. Make this shot night. Give her a coat. Cut two seconds here. The machine changes that one thing without re-rolling the rest.

The pitch is pointedly control first. It is not one promptPromptThe text description you provide to a model to specify what you want it to generate. and a finished film, it is a director giving notes to a crew that happens to be made of models. That framing matters, because the loud version of AI film, type a sentence and receive a movie, is the version that does not work, and the quiet version, hold a vision and correct the output, is the one that does.

This is a real shift in where the human sits. Across the workflow chapters we watched the interface climb from the model to the aggregatorAggregatorA tool or platform that wraps many image and video models behind one interface, so you switch models without switching apps. to the agent. Vibe directing is what that climb looks like from the creator's chair. You stop operating the tools, stop wiring the pipeline by hand, and start doing the thing a director actually does, which is hold a vision and give notes until the thing on the screen matches it.

It is worth being precise about what is new and what is not. Generating a clipCLIPA text encoder developed by OpenAI in 2021 that learns to align text and images in a shared embedding space. Foundation of most text-to-image models from 2022 onward. from a prompt is old. What is new is the division of labor. The system now carries the shot-to-shot consistency, the audio, the cuts, and the assembly, and the person carries taste and intent. The scarce skill stops being who can drive the software and becomes who has something to say.