How I work with AI
Everything below is the part most studios leave vague. AI is the production engine here — not a secret, and not the selling point. The story directs every decision; the tools execute it.
The machine renders. It does not decide.
Generative tools are very good at producing an image and very bad at knowing which image is worth producing. Every judgement that determines whether the work lands is made before a model is ever opened, and remade every time one returns something that is technically correct and strategically useless.
The split is simple enough to state plainly:
- Read the brief and find the actual problem
- Write the narrative and the positioning
- Decide the concept, the world and the tone
- Cast, frame, light and sequence
- Reject the eighty per cent that is nearly right
- Cut, grade, sound and finish
- Render frames from a direction already set
- Produce variants fast enough to compare
- Extend, clean and retouch plates
- Voice a script that was written first
- Make iteration cheap enough to be honest about what isn't working
Direction, not prompts.
Anyone can reach the same models for the price of a subscription. That is genuinely true and worth saying out loud — it is also why the prompt is not the product.
What you're paying for is the judgement that decides what to make, the consistency that holds a campaign together across forty assets instead of four good ones, and the editorial discipline to throw away the output that is impressive but off-brief.
The background matters here: brand strategy and copywriting first, production second. The tools came last, and they came to serve the first two.
What this workflow is bad at.
A studio that lists no limitations is selling something. These are real, and they should shape whether you hire me for a given job.
Exact product fidelity
If your product has precise geometry, legible packaging copy or a logo that must be pixel-correct, generation alone will not hold it. That work needs real product photography as a base, or careful compositing on top. I'll tell you which at the brief stage rather than at delivery.
A specific real person
A named founder, a signed ambassador, an existing spokesperson. That is a shoot, not a generation, and pretending otherwise creates both a likeness problem and a quality one.
Documentary truth
Anything that has to be evidence of a real event — a real customer, a real location, a real result — cannot be synthesised. Not a technical limit, a truthfulness one.
Long-form continuity
Character and set consistency holds well across a campaign of short pieces. Across several minutes of continuous narrative it degrades, and the honest answer is that the work gets shaped around that rather than fighting it.
The position, in plain terms.
No real likeness without a release
Generated figures are synthetic and are not built from, or intended to resemble, a specific identifiable individual. Where a project needs a real person, that becomes a casting and release conversation before anything is produced.
Synthetic creative is labelled as such
UGC-style work is synthetic advertising creative, not genuine customer or creator content. Presenting it as a real testimonial would be a false claim however it was produced, and I won't help do it. How you disclose beyond that is your decision — platform rules on Meta, TikTok and Google currently require some form of it, and they change often.
Ownership
Copyright in purely AI-generated images is unsettled in several jurisdictions, including the EU and the US, where protection generally turns on demonstrable human authorship. What I can transfer is every right I hold, and I'll put the specifics in writing before a project starts. Full terms are on the FAQ.
Your material stays yours
Brand assets, product references and unreleased material you send are used for your project and nothing else — not published, not shown as work in progress, not used to train anything.
This portfolio is spec work.
Almost everything on this site is self-initiated. The brands are invented, the briefs are my own, and the campaigns never ran. They are speculative examples and they carry no performance claims — no view counts, no conversion lifts, no case-study numbers, because there aren't any to report.
That is deliberate rather than apologetic. Spec work is where you can see the direction clearly, without a client's compromises layered over it. What it can't show you is how I handle feedback, deadlines and a stakeholder who changes their mind in week three.
The closest thing to an outside brief is KROMA, a team-developed Master's capstone — verbal identity, copy system, campaign assets, video ads and a Shopify prototype. It is still speculative: no clothing was made, no certifications were obtained and no campaign ran. But the brief came from outside and was taken to completion under academic deadlines.
Still have a question about it?
The awkward ones are the useful ones. Ask them before we start rather than after.