Automating your visuals with AI: brand, law, method

Yes, you can automate part of your visual production. Social media visuals, article illustrations, format and language variations: generative models now produce output that is usable in a business setting. The caveat can be summed up in three words — brand, law, data. Without a framework covering these three points, automation ends up costing more than it saves.

Article published in January 2026, revised on 25 May 2026, reviewed on 7 July 2026.

Which visuals does AI produce correctly in 2026?

In 2026, image generation reliably covers the everyday visuals an SME needs — social media posts, article illustrations, format and language variations — but it does not replace brand-identity photography or art direction.

The market isn't reducible to a single product, and I'll deliberately avoid naming any: products change every six months, categories endure. Here are the four tool profiles I distinguish, along with their main trade-off.

In short: no profile dominates. The right choice depends on your volume, the sensitivity of what you transmit, and the skills available in-house — not on whichever tool everyone is talking about this quarter.

The first real risk: your brand drifting

The main gap between experimentation and sustainable use isn't about the quality of a single visual. It's about consistency over time. Every generation produces a slightly different result; without a framework, colours drift, compositions vary with no logic, and the style stops being recognizable. This drift is invisible in a single image and glaring across three months of a content feed.

What keeps consistency intact is well understood. First, a documented visual framework — often called a brand prompt: expected style, dominant palette, recurring elements, subjects to avoid, overall tone, shared by everyone who generates visuals. For organizations with the technical capacity, further training an open model on your brand guidelines then makes generations naturally consistent, without heavy phrasing needed for every image. And in every case, systematic human validation: no image goes to publication without review by a visual point person, because models still produce six-fingered hands, illegible text, and spatial inconsistencies.

What does Swiss law say about a generated image?

Three elements of Swiss law apply directly, and none of them is theoretical.

First, the Copyright Act^[1]. A work must result from intellectual creation to be protected. A fully generated visual, without significant human creative input, may not benefit from this protection — the assessment is made case by case. Concrete consequence: if your entire visual output comes from a generator with no documented creative work, you might be unable to stop a competitor from reusing it.

Then the Unfair Competition Act^[2]. A misleading visual — a fake customer testimonial, a staged scene presented as real, a fabricated professional endorsement — can amount to a sanctionable unfair practice. The rule doesn't target the technology; it targets deception, regardless of how the image was produced.

Finally, the right to one's own image. Reproducing the face of an identifiable person without their consent remains illegal, whether the image was photographed or generated. This is probably the most immediate risk area: a model can unintentionally produce a resemblance to a real person that you won't even recognize as such.

For companies active on the European market, the European AI Act is progressively adding transparency obligations regarding the origin of generated content^[3]. Better to anticipate them than to discover them the day they become binding.

Where does your data go when you generate online?

An online tool processes your images and descriptions at the provider's end — that's inherent to how it's built. Before uploading anything, qualify what you're transmitting. Personal data in your briefs: describing identifiable customers or employees constitutes processing under the Federal Data Protection Act, and requires that you've reviewed the provider's terms. Intellectual property in your references: the photograph you upload to get a variation may be protected by a third party's rights. And finally, strategic confidentiality: an unlaunched product, a campaign in preparation, or an identity being redesigned have no business being sent to a provider without an explicit contractual commitment.

When these risks are substantial, a custom solution built on an open model and run on infrastructure you control resolves the issue structurally — at the cost of requiring more demanding in-house expertise. It's the same underlying trade-off I describe in SaaS versus custom-built: convenience online versus control on your own premises.

Do you still need a graphic designer? Yes — elsewhere.

Automation doesn't eliminate visual expertise; it shifts where it's applied. Variations, adaptations, and language localizations get absorbed by the tools, provided there's a clear framework. What rises in value in return: art direction, because without it automated production converges toward a sameness that differentiates no one; visual strategy, which decides where generation is appropriate and where human expertise remains structurally essential; quality oversight, which becomes a role in its own right, one to structure rather than scatter.

For an SME without an in-house designer, this setup works with an external point person — an independent graphic designer, for example — who owns art direction and quality oversight while your teams produce the everyday visuals.

How to get started without damaging your image

Start with a specific use case — social media visuals or blog illustrations are good entry points — rather than attempting to automate all of your production at once. Train at least one person in prompt framing, document your visual framework, and build a library of proven phrasings. And combine approaches: generate the first draft with a model, then refine it through retouching to reach a professional standard. This combination recognizes the value of the tools without overestimating them.

A word for sectors where authenticity matters most — luxury, gastronomy, tourism: an exclusively generated visual language carries a positioning risk that your customers perceive, sometimes without being able to name it. Real photography remains your proof; generation, your volume.

Finally, frame this initiative like any other automation project: the workflow first, the tool second. And to place visual production among an SME's other AI initiatives, the broader framework is laid out in What should a Swiss SME do about AI in 2026?

Key takeaways

— Image generation covers everyday visuals; brand identity and art direction remain human work. — Without a documented visual framework and pre-publication validation, your brand drifts image after image. — A fully generated image may be impossible to protect under Swiss copyright law; the right to one's own image and the Unfair Competition Act, however, apply in full.

FAQ

Do I own an AI-generated image? Usage rights depend on the provider's terms, but protection under Swiss copyright law is uncertain: without significant human creative input, the image may not be protected for anyone. Document your creative work — prompt framing, retouching, composition — if you want to be able to defend your visuals. The assessment is made case by case.

Do I need to disclose that an image is AI-generated? The line to remember: the Unfair Competition Act sanctions deception, not technology. A staged scene presented as real is a problem; an obvious illustration is not. For the European market, the AI Act is progressively adding transparency obligations for generated content — anticipate them if you operate there.

Can I upload photos of customers or employees to an online tool? Not without prior assessment. These images contain personal data, and transmitting them to the provider constitutes processing under the Federal Data Protection Act: review its terms and your basis for consent before sending anything. For sensitive content, infrastructure you control remains the structural answer.

What budget should I plan for? Online tools are generally billed through modest monthly subscriptions, on the order of a few dozen francs per user; that's the visible part. The real investment lies in the framework: documenting your visual guidelines, training a point person, quality oversight. Controlled infrastructure costs considerably more upfront and is weighed against the sensitivity of your content.

What about your own visual production? The AI Usage Diagnostic: sixty minutes to map your actual workflows, identify what deserves a custom build, what should stay in SaaS, and what doesn't need AI at all. Book a diagnostic

Sources

[1] Federal Act on Copyright and Related Rights (CopA), of 9 October 1992. www.fedlex.admin.ch/eli/cc/1993/1798_1798_1798/fr [↩]

[2] Federal Act against Unfair Competition (UCA), of 19 December 1986. www.fedlex.admin.ch/eli/cc/1988/223_223_223/fr [↩]

[3] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence. eur-lex.europa.eu/eli/reg/2024/1689/oj/eng [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.