What an AI Image Generator Shows When You Cannot Show the Work

Count the projects you have shipped in the last three years. Now count how many you can actually put online.

For most working designers the second number is small and sometimes zero.

Internal tools nobody outside the company will ever see. Clients work under a contract that was signed before anybody thought about portfolios. A product that got built, got good, and got shut down in a restructure. White-label systems where the whole point is that your name is not on it.

So the portfolio shows three things from four years ago and a side project, which is a poor representation of somebody who has been busy and competent the entire time.

An AI Image Generator, integrated into Higgsfield, helps with part of this. It also makes one specific mistake considerably easier to make, and that gets covered before anything else.

Why is most of your work unshowable?

Four reasons, and most designers have hit all of them.

Confidentiality. Standard in client contracts and frequently broader than people realise, covering not just the work but the fact of the engagement.

Internal products. Nobody outside the organisation has seen the admin panel you spent eight months on, and nobody is going to.

Unlaunched work. The thing that was designed, approved, built halfway and then shelved. Technically yours, practically invisible, and often the strongest work you did that year.

And employer policy, which may prohibit publishing anything from your time there regardless of what the contract says about ownership.

None of this is unusual. It is simply the normal condition of working in design for anybody not freelancing exclusively for consumer startups who love publicity.

What do designers do about it now?

Three things, and one of them is a problem.

Some show nothing and let the portfolio go stale. The work exists, nobody can see it, and the next application is harder than it should be.

Some show the same two public projects indefinitely, which signals less experience than they actually have.

And some post things they should not, which is the one worth being direct about. Screenshots of internal tools. Client work before launch. Redacted interfaces that are still recognisably the client's product.

That third group is not being reckless. They are solving a real problem with the only material they have, and the risk sits quietly until somebody from the client's team sees it on Dribbble.

There is a fourth option and it is underused, which is most of what this article is about.

What can you legitimately show?

More than people assume, once you separate the work from the artefact.

The problem, described generically. Not the client, not the product, but the class of problem. Reconciling two data sources with conflicting formats. Onboarding somebody into a workflow with eleven steps.

Your approach and reasoning, which is the part a hiring manager actually wants and the part that is rarely confidential.

Your own exploration, including directions that were not chosen, which frequently belongs to you in a way the final shipped design does not.

Abstracted representations of structure. A flow, a hierarchy, an information architecture, stripped of anything identifying.

And outcomes where you are permitted to state them, which varies entirely by contract and is worth checking rather than assuming.

What you cannot show is the actual interface, the client's identity, their data, or anything that lets somebody recognise the product.

Where does an AI Image Generator fit?

Producing the visual layer for a case study that has no screenshots in it.

A written case study with no images performs badly. That is not a comment on the writing, it is how portfolio sites and recruiters work. Something has to be on the page.

An AI Image Generator gives you material that is clearly illustrative rather than documentary. Abstract representations of structure, visual metaphors for the problem, context imagery that sets the scene without depicting anything real.

The specific thing an AI Image Generator does well here is abstraction. A problem about fragmented data has a shape, and describing that shape produces something usable, where a stock library gives you a photograph of a laptop.

An AI Image Generator also produces diagrams, which are the strongest asset in this situation. A flow with generic labels showing how you restructured a process says a great deal about your thinking and nothing about the client.

Higgsfield holds that work in one place as an AI creative suite, which matters mainly because a case study needs several pieces at different sizes and the alternative is three tools and a resize.

What must never be presented as the real thing?

A generated interface shown as a product you designed. This is the line and it is absolute.

It is easy to produce something that looks like an app screen. It is also fabricating a credential, because a hiring manager looking at it believes they are seeing your work.

The consequences are not hypothetical. Design interviews involve walking through your own case studies in detail, and being asked about a screen you did not make goes badly within about two questions.

Never use an AI Image Generator to approximate the client's actual product, even approximately. If it is recognisable, the confidentiality problem you were avoiding has arrived anyway.

Never generate data, metrics or results in an AI Image Generator. If you cannot state an outcome, say the outcome is confidential, which reads as professional.

Never generate a logo, a brand or anything identifying.

The working test is whether somebody could mistake the image for a screenshot. If yes, it is in the wrong register. These should be obviously illustrative.

Is a blurred screenshot safe?

Usually not, and this catches people out.

Blurring the text leaves the layout, the navigation structure, the brand colours and the overall product shape entirely intact. Anybody who has used the product recognises it immediately, and anybody at the client company recognises it instantly.

It also signals that you had something you should not be showing, which is a worse impression than showing nothing.

Heavy pixelation has the same problem plus a technical one, since some implementations have been reversed under certain conditions.

Abstraction is different from redaction. A redacted screenshot is the real thing with parts removed. An abstract diagram is a different artefact that describes structure without reproducing it.

The second is defensible. The first is the same disclosure with a filter over it.

What about showing process instead of product?

The strongest option available and the one most designers skip.

Process artefacts are frequently less restricted than final designs, because the confidentiality concern is usually about the product and the client rather than about how you think.

Sketches, flows, decision trees, the three directions you explored and why two were rejected. All of that demonstrates capability and most of it can be abstracted without losing meaning.

It is also what senior roles are assessed on. A portfolio of polished final screens shows execution. A case study showing how a problem was framed, what was tried, and why the chosen direction won shows judgment, and judgment is what gets people hired above a certain level.

An AI Image Generator renders those artefacts cleanly, which matters because the versions in your notebook are not presentable and nobody has time to redraw them properly.

Check your contract before publishing process material too. Some confidentiality clauses cover everything produced during an engagement rather than the deliverable alone.

What does this do for the work you cannot name at all?

A harder case, and more common than the standard NDA situation.

Some engagements prohibit naming the client entirely. Not just the product, the fact that you worked with them. Government contracts, acquisitions in progress, competitive situations, anything under an agreement written by somebody cautious.

That rules out most case study structures, because the usual format opens by saying who it was for.

The version that works reframes around the problem class instead. A case study titled around reconciling conflicting data sources, or around onboarding into a multi-step workflow, carries no identifying information and still demonstrates exactly what you can do.

Sector can be stated broadly if the contract allows it. Financial services rather than a named bank. Healthcare rather than a named provider. Check first, because in a small market even the sector plus the timeframe narrows it to three companies.

An AI Image Generator earns its place here more than anywhere, since the case study has no client, no product name and no screenshots. Everything visual has to come from the structure of the problem, which is exactly what abstraction is for.

Higgsfield keeping those pieces organised by case study matters when you have several, because the images for an unnamed project are hard to identify from a filename six months later.

And the writing carries the most weight in this version. If the reader cannot be told who or what, the description of how you thought has to do all of it.

Does a case study need images at all?

It needs something, and the bar is lower than designers assume.

A case study that is pure text gets skimmed and abandoned. Portfolio reviewers move quickly and a wall of paragraphs reads as effort they have not committed to yet.

What breaks that is any AI Image Generator structure on the page. A diagram, an abstracted flow, section imagery, anything that gives the eye somewhere to rest.

Three or four pieces per case study is enough. One setting the context, one or two showing structure or process, one closing.

That is considerably less than a conventional case study full of screens, which is fortunate, because you do not have the screens.

And the writing carries more weight here than in a normal portfolio piece. If the images cannot show the work, the text has to describe it well, which is a skill worth developing regardless.

What should you check before posting?

Four things, and they take about twenty minutes once.

Your contract or employment agreement, specifically the confidentiality clause and anything covering publication. Many designers have never read theirs.

Whether the engagement itself is confidential. Some contracts prohibit naming the client at all, which changes how the case study is framed.

Whether anything you are describing is still unreleased, since timing clauses are common and frequently expire.

And whether your abstraction is actually abstract. Show it to somebody unfamiliar with the project and ask what product it is. If they guess correctly, go further.

If any of it is unclear, ask. A short message to a former client or manager asking what you may publish is a normal request and is answered positively more often than designers expect.

What would one case study take?

An evening, for a project you remember well.

Write it first, before opening Higgsfield, with no images at all. The problem, your approach, what you tried, what was chosen, what happened. Around five hundred words.

Strip out everything identifying. Client, product name, sector where it narrows things too far, specific figures you cannot support publicly.

Produce three or four pieces in an AI Image Generator. One for context, one or two abstracted diagrams of the structure or flow, one to close.

Keep the Higgsfield output obviously illustrative. If any of them could be mistaken for a screenshot, change it.

Save the treatment so the next case study matches. Higgsfield storing it means the second one takes a fraction of the time, which is what determines whether you write a third.

Then show the whole thing to somebody who was not involved and ask them what product it is.

FAQ's

Can an AI Image Generator recreate a designed interface?

No, and it would be fabricating a credential. Anything presented as your work has to be your work.

Is a blurred screenshot acceptable?

Rarely. Blurring leaves the layout, structure and brand intact, and it signals you had something you should not be posting.

What can be shown from NDA work?

The problem abstracted, your approach and reasoning, your own exploration, and structural diagrams with generic labels. Check your contract first.

How many AI Image Generator pieces does a case study need?

Three or four. One for context, one or two showing structure, one closing. Less than a conventional case study, which is convenient.

Is process work less restricted?

Often, though not always. Some confidentiality clauses cover everything produced during the engagement rather than the deliverable.

Is there a free way to try this?

Higgsfield has a free tier, which covers one case study's worth of images.

Conclusion

A portfolio that shows three projects from four years ago does not represent somebody who has been working steadily the entire time. It represents the small fraction of their work that happened to be publishable.

An AI Image Generator fills the visual layer for case studies built on abstraction rather than screenshots, and Higgsfield keeping the treatment consistent means writing the second one is not a fresh start.

What it cannot do is show work you are not allowed to show. If an image could pass for a screenshot of something real, it is the wrong image, and the version of this that goes wrong always goes wrong in the same direction.