If a creative today wants something specific from a generative AI model — something that was never in the training data, because it never existed — they can go one of two ways. Bend the vision until it matches what the model happens to produce, and settle for mediocrity. Or walk away from the tools entirely.

We choose the third way.

The decisions are made before the model runs.

We created a deterministic layer between the creative team and the model: a set of protocols and practices that define, document and quantify every creative decision before the model runs. Those decisions are then passed to the model as fixed parameters. That way we use the model as a finishing coat over a creative vision ultimately forged by people.

What does this mean in practice?

• If we want to make a character look a certain way consistently — we can.
• If we want a believable, authentic emotional transition from a character at a precise moment in time — we can.
• If we want the character to have a specific behaviour and move in a certain way through a defined environment — we can.
• If we want the camera to move on a set trajectory relative to the character or the environment, with a lens at a given focal length and focus pulled on a chosen beat — we can.

You see the real thing before you commit to it.

Because the shot is built before it is generated, we can show the actual blocking, lens and camera move within days of a brief. The model then generates from a very precise prompt, with many of the variables fixed and no internal contradictions, which means far fewer shot-in-the-dark generations and faster results.

The job defines the tool, never the reverse.

Our method is model-agnostic. Whenever a new model comes out, we can simply swap it into our system. Beyond that, our philosophy is that the job defines the tool, never the reverse, and different models suit different parts of the process.

Sometimes a fully AI-generated piece is the right answer. Sometimes it is a hybrid: a more traditional filmmaking approach with actors, a camera and a crew, with VFX and generative work carrying what a camera cannot reach. That is the process where the collaborative decision-making of a director, working with actors, a DOP and a production designer, can put a bolder vision on screen.

Made so it can actually run.

We follow the most current best practices and protocols designed to minimise legal risk for us and our clients, in an environment that is still legally uncertain. Before we make anything, we check the licences of the tools against the markets the work will run in, so the way a piece is made never compromises how it is distributed or who owns it. We have turned part of how we do this into a free tool that anyone can use on their own stack. You can test it right here.

Alongside every production we keep the record that makes the work defensible:

• Raw renders archived and hashed before anything touches them.
• A per-shot log of tool, version, account, inputs and rights, written as it happens.
• Frame-level clearance by a named person, dated.
• The grade kept non-destructive beside the original.
• The EU AI Act disclosure designed at treatment stage instead of bolted on afterwards.