← Back to Blog

AI-assisted creative work: from brief to finished assets

Abstract illustration of design panels and colour swatches arranged along a production line

Most creative teams now use AI for something. Far fewer have turned it into a dependable way of producing finished work. There's a big difference between trying out an image generator for mood boards and delivering a full set of on-brand, accessible assets that a client can approve and use the same week.

This post explains how we use AI in our own creative production: the stages of a project, the tools involved, where AI-generated work tends to fall down, and how we deal with that. It's written for business owners commissioning creative work as much as for other agencies.

Three levels of AI use

It helps to think of AI adoption in three broad levels. Where a team sits largely decides whether AI saves real time or just adds another tool to pay for.

Level 1: ideas

Designers use ChatGPT or Claude for copy ideas, an image model for mood boards, and Adobe Firefly for quick variations. It's useful, but it's a set of separate tools rather than a process. Projects start faster, but the middle and the end take as long as they always did.

Level 2: joined-up automation

Here the tools are connected. A brief sets off a sequence in which concepts are generated, refined, converted to every size and format needed, and sent for review. The dull production work, such as resizing, converting and organising files, happens without anyone doing it by hand, and designers spend their time on decisions.

Level 3: agents with oversight

At the leading edge, AI agents can operate design software directly, and brand rules are enforced at the moment something is generated. This is where we work on production jobs, but it only works with the right set-up and with people checking the output at the right points.

Five-stage creative workflow: brief and prompt planning, generating concepts and variations in parallel, technical refinement for brand and accessibility, a client review package, and handover with source files and prompt logs
The five stages of an AI-assisted creative project.

How a project runs

The point isn't speed for its own sake. It's removing the waiting between decisions: the hand-offs, the manual stitching together and the "I'll get to that tomorrow" that stretches two days of work into three weeks.

1. Brief and prompt planning

We break the client's brief down into the things a generator needs to know: style, brand rules, audience and what success looks like. The old rule still applies: vague input produces vague output. A good brief at this stage is the difference between fifty usable variations and five hundred unusable ones.

2. Generating concepts and variations

Several streams run side by side: image models for visual concepts, a language model for copy, and Firefly where licensing needs to be clean. The volume gives the person choosing a real choice rather than "this one or that one".

3. Technical refinement

This is where most AI work falls apart. Generating concepts is easy; making them ready for production is not. Every asset is checked for brand consistency (colours, type, logo placement and spacing), accessibility (contrast, alt text), the right sizes and file types for each platform, and licensing.

4. Client review

The client gets one tidy review space with numbered versions, notes on the design decisions and an easy way to approve or comment. Nobody has to ask which version they're looking at.

5. Handover

At the end the client receives the source files, the brand rules we worked to, the prompt log so they can produce more variations themselves, and the accessibility results.

Where AI creative work goes wrong

Most teams experimenting with AI hit the same wall. The concepts look impressive, but turning them into assets a client can actually use is much harder. The usual problems are these.

Brand drift

AI tools know nothing about your brand. Left alone, each result is a gamble. We turn colours, type scales, logo rules and visual style into constraints the generators have to work within, so the output stays recognisably yours.

Accessibility

UK businesses have duties under the Equality Act 2010, and those selling into the EU may also fall under the European Accessibility Act. AI-generated images regularly fail on contrast and arrive with no useful alt text. We check contrast on every asset and write and review alt text before anything reaches the client.

Version chaos

Generate eighty concepts in an hour and keeping track becomes a job in itself. We keep creative assets under version control, with a separate branch for each round of feedback, so nothing gets lost or overwritten.

A poor handover

The worst thing a client can receive is a folder of files with no explanation. Ours comes with the context needed to use, change and reuse the work.

The tools we use

Tools change quickly, so treat this as a snapshot rather than a recommendation. For ideas and copy we use the leading language models, with image models for visual concepts and Adobe Firefly where commercially safe imagery matters. Automation runs on a self-hosted n8n instance on UK servers, because clients' work and brand assets shouldn't sit in someone else's cloud by default. Production happens in Figma and Adobe Creative Cloud, with a code editor for email templates and prototypes. Finished assets go through a pipeline that optimises them and delivers them to the client's review space.

What it means for cost

AI-assisted production usually takes far fewer hours than the traditional approach, especially on high-volume work such as social media variants or resizing a campaign for every platform. That saving can be passed on to the client. What it doesn't remove is the need for skilled people to set the direction, check the quality and take responsibility for the result.

Frequently Asked Questions

How much faster are AI-assisted creative workflows?

On high-volume production work, such as variants, resizing and format conversion, the time saved can be dramatic, turning days into hours. On the strategic and creative decisions, the gain is smaller, because that part still needs people.

Can AI replace human designers?

No. AI handles much of the repetitive production, but the parts that decide quality, such as the idea, brand judgement, cultural sensitivity and the client relationship, still need experienced people.

How do you keep AI-generated assets on brand?

By turning the brand guidelines into rules the generators must follow, checking every asset against the design system, and having a person approve the final set.

Are AI-generated images safe to use commercially?

It depends on the tool and its licence terms. Where licensing matters, we use tools trained on licensed content, such as Adobe Firefly, and check the terms for each project.

Want help with creative production?

Tell us what you need and we will explain how we would approach it, with a fixed price.

Compare the plans →