Exploration at volume
Generate many directions quickly so the choice is made against real options rather than one safe idea. The generating is fast; the choosing is still a designer's job.
AI design services, with a designer still accountable.
Norml uses generative tools in real design work — exploration, asset production, variants, first drafts — and does not use them for the decisions that need judgement. This page is the honest version of where the line sits, because it is the question clients ask before they hire us.
A few clients we’ve worked with
AI made producing design output cheap. It did not make deciding cheap.
Output went up and the work stopped looking like one company.
A generated identity looks competent and could belong to any business in the category.
Nobody can say whether the imagery is licensed, and it is already on the site.
A model produced a layout that looks finished and fails the moment real content goes in.
The team wants the speed and has no rule about when generating is and is not appropriate.
What our AI design services include.
We agree up front which parts of the work are generated, which are drawn, and who signs off. The speed is real, and it comes from spending it on exploration and production rather than on decisions.
Asset and imagery production
Illustration, icon sets, textures and supporting imagery produced to a defined style, with provenance and licensing recorded per asset.
Variant production from a system
Once the design system is settled, produce the long tail of states, sizes and page variants against it instead of by hand.
Design systems AI can work inside
Tokens, components and naming documented so both your team and a model produce output that is consistent rather than merely plausible.
Review against an anti-slop standard
A written checklist of the patterns that mark generated design — the same five palettes, the centred everything, the decorative numbering that encodes nothing — checked before anything ships.
A usage standard for your team
What may be generated, what must be drawn, what must be licensed, and what must never be automated, written down so it survives without us.
AI design deliverables.
The agreed scope identifies which of these deliverables your project needs.
- 01
Explored directions with the reasoning for the one chosen.
- 02
Produced assets with provenance and licensing recorded for each.
- 03
A design system documented so generated output stays consistent.
- 04
A review pass against the anti-slop checklist, with findings written down.
- 05
A written AI design standard your team can apply themselves.
Our process
How our AI design process works.
Agree the line
Decide together what is generated and what is drawn, before any work starts. This conversation prevents most of the disappointment.
[ 01 ]
Explore
Many directions, fast, so the decision has real options behind it.
[ 02 ]
Decide
A designer chooses and justifies. This step is not automated.
[ 03 ]
Produce
Assets and variants against the agreed system, with provenance logged.
[ 04 ]
Review and hand over
The anti-slop pass, then the files, the system and the written standard.
[ 05 ]AI design questions.
What are AI design services, exactly?
Design work where generative tools are part of the production process: exploring directions, producing assets, generating variants from a settled system. It is not a tool you use yourself, and it is not a generated logo. If you want a free generator for a room or a logo, that is a different product and plenty of sites sell it.
Will our brand look AI-generated?
That is the risk, and it is why the review step exists. Generated design clusters around a handful of recognisable looks — the same warm cream and terracotta, the same acid accent on near-black, everything centred, decorative numbering that encodes nothing. We keep a written list of those patterns and check against it before anything ships.
Who owns the output, and is the imagery licensed?
You own it, and we record provenance per asset so you can answer that question later. Anything with unclear rights does not go in. This matters more than it seems: the expensive version of this problem arrives long after launch.
Does this make design cheaper?
It makes parts of it faster — exploration, assets, variants — and the fast parts were never the expensive parts. What it does not compress is deciding what the thing should be, which is most of the value. We would rather say that than quote a discount we cannot honour.
Is this the same as designing a website for an AI company?
No. That is a different page: designing and building the site of a company whose business is AI. This one is about AI being in our own production process.
Can you set a standard our in-house designers follow?
Yes, and it is often the most useful thing here. A written rule for what may be generated, what must be drawn, what must be licensed, and what must never be automated.
Website 2.0 combines strategy, design, development and launch in one engagement, with the AI line agreed before work starts.