How we use AI

AI widens and accelerates. It covers more sources in research, produces more directions at the start, generates variations once a direction is set, and takes the routine out of production and code. What it does not do is decide. The problem a project solves, the positioning, the concept, the visual language and the final quality are all judgements made by people, and we own them. There is a full breakdown by stage above: for each phase of work it says what the tools carry and where the decision stays with us.

We work on paid business plans where inputs are not used to train the models, and we check that setting for every tool we add to the stack. Anything under NDA stays inside the tools we have agreed with you, and sensitive material (personal data, client lists, contracts, credentials) does not go into generative tools at all. If your security or compliance team has its own requirements, send them to us before we start and we will work to them, or tell you plainly which part of the process would have to change.

Yes. We can run a project without generative tools, or keep them out of specific assets, for example anything customer-facing or anything your legal team needs clean. It changes the timeline rather than the outcome: some stages take longer, and we will tell you which and by how much before we start, so the decision is made with the real trade-off in front of you, not after the fact.

One test: does this step need coverage or judgement? Where the job is to get through volume (sources, variations, storyboards, repetitive code), the tools help and we use them. Where the job is to decide what something means and what it should look like, a person does it. The line sits in roughly the same place on every project, and where a client’s rules or the nature of the material move it, we say so at the start rather than mid-way.

By treating generated material as raw footage, not as output. On the WHOOP campaign a single scene took up to twenty takes before one was right, and everything that survived was graded, retouched and detailed by hand. There are specific habits behind it: correcting the magenta cast that generators push, regenerating a failed detail separately in 4K and merging it in, keeping clips short enough to control. The reason the work does not look generic, though, is earlier than any of that. The concept and the visual language are ours, so the tools are executing a direction they could not have arrived at.

The rights to the final work are yours, as they would be on any project. On our side, AI is a tool inside the process, not a way around authorship. References and moodboards set direction and are never reproduced. We do not prompt in the style of a living artist, and we keep other brands, characters and recognisable people out of generated material: the heroine in our WHOOP campaign is a synthetic character, not anyone’s likeness. Everything generated is reworked, composed and finished by hand before it becomes a deliverable.

Faster, yes, and in specific places: research, early exploration, production of series, routine code. Cheaper is less straightforward. The time saved on volume goes back into selection, correction and finishing, which is where quality is actually decided. What you get for the same budget is more explored before a direction is chosen, and more produced once it is, rather than a smaller invoice for the same work.