Three years ago, asking an AI for an image got you six-fingered hands, warped faces and backgrounds that made no physical sense whatsoever. Today Midjourney, DALL-E 3 and Adobe Firefly produce images that, at first glance, could pass for a professional stock photo shoot. That leap has triggered the question we hear every week from small business owners: "why do I need a photographer or a designer when I can generate my company's photos in two minutes for free?" The short answer: for quite a bit less than you thought a year ago, but still for more than you'd like to hear.
It's worth separating the hype from the actual use case. AI image generation has moved from being a curious toy to being a genuine marketing tool, but like any tool, it has a territory where it shines and one where it still fails quietly (and that quiet kind of failure does the most damage, because you don't notice it until the image is already published).
Where AI image generation genuinely works today
There are uses where these tools aren't an imperfect substitute for a professional: they're simply the faster, cheaper option, without any real loss of quality. The first is ideation. When you need to explore ten different visual directions for a campaign before deciding which one deserves an actual photo shoot, generating those ten variations with AI in half an hour saves you days of sketches, moodboards and loose references pulled from Pinterest. This used to be the phase that burned the most time without producing anything tangible, and now it gets resolved in an afternoon.
The second is product mockups. If you sell something physical and need to see how your logo looks on a mug, a paper bag, a t-shirt or a billboard before printing anything, AI generates those mockups instantly at zero production cost. This used to require an expensive Photoshop template or a designer spending half a day compositing the image layer by layer.
The third is backgrounds and settings. A product photo shot against a neutral background in a studio (something still worth doing with an actual camera, not AI) can afterwards be placed in a kitchen, a bright office or a generated landscape, without renting a location or building a set. Home decor and fashion brands have been doing this routinely for over a year now for their online catalogues, combining real product photography with generated backdrops that change from campaign to campaign.
And the fourth use, less talked about but very profitable, is bulk product variation. If you carry 40 references of the same t-shirt model in 12 different colours, generating colour variations from a single base photograph is far cheaper than photographing all 40 combinations one by one. Here AI doesn't replace the original photo (that's still needed as the base), but it multiplies its output without multiplying the cost.
A practical case: a home decor catalogue
Picture an online store with 60 home decor products: vases, cushions, mirrors, lamps. Booking a full photo shoot with styling, location and several scenes per product can cost several thousand euros and take weeks to deliver. An increasingly common mixed approach is to photograph each product once, with good lighting against a neutral background (a one-day shoot, much more affordable), and then generate five or six different settings per product with AI: a Nordic living room, a Mediterranean one, a minimalist one. The result is a catalogue with genuine visual variety for a fraction of the cost and time it would take to shoot each product in each physical setting. The key to making this work isn't the AI itself, it's never losing quality control: every generated image gets reviewed before publishing, because a vase that "floats" slightly above the table or a shadow that doesn't match the scene's lighting destroys buyer trust in under a second.
What still shows, even if less often
The old joke about six-fingered hands barely applies anymore: current models get hands right most of the time. But other, less obvious and more dangerous failures for a brand still persist. Text inside the image (a sign, a product label, a background poster) still comes out as letters that don't form real words in a high percentage of generations. Symmetry on repeated objects also fails often: a row of chairs where one has an extra leg, a bookshelf where the books morph into impossible shapes toward the back of the image. And reflections or shadows don't always match the scene's actual light source, a detail the human eye catches instinctively even without being able to explain why "something looks off".
The more serious problem for a business, though, isn't technical: it's brand consistency. Generative AI doesn't "know" your exact corporate colour palette, your logo's typography, or how a model should pose to convey what your brand wants to convey. You can ask it for "a photo in my brand's style" and you'll get something generic that resembles a thousand other brands, not yours. Keeping a recognisable visual identity across a hundred AI-generated pieces takes real art direction work (choosing references carefully, reusing the same base prompt, manually adjusting each result) that the machine doesn't do on its own, no matter how good the model behind it is.
Image rights: the part almost nobody checks before publishing
This is where most companies get into trouble without realising it. AI models were trained on billions of images from the internet, many of them copyrighted, and that origin leaves traces: there are documented cases of generated images that reproduce, almost literally, the style, composition or even recognisable elements of existing photographs and illustrations. Using that image in a commercial campaign exposes you to a claim, even if you had no idea the original existed.
On top of that, legal ownership of an AI-generated image isn't settled the same way in every country or for every use. For a serious commercial use (a paid ad, product packaging, a wide-reach campaign) it's worth using tools with clear commercial licensing (Adobe Firefly, for instance, is trained only on licensed or public-domain content, which cuts the risk considerably) and keeping a record of which tool and which licence covered each image you publish. For lower-risk content, like a supporting image on a blog post or a moodboard idea that never leaves the office, the real risk margin is much smaller.
There's a third point that gets forgotten even more often: if you generate the image of a real person (or something very close to it) without their consent, you're in the territory of personality rights, which in most jurisdictions, including Spain, are protected separately from copyright and can cause problems even if the image is technically "original" and machine-made.
When you still need a real photographer or designer
There are three situations where it's not worth gambling on generative AI today. The first is any image where the person shown is you, your team or your actual clients: nobody trusts a brand whose "team" on the website turns out to be people who don't exist, and once it's discovered (and it gets discovered, because people are getting better at spotting these patterns) the credibility damage far outweighs the savings. The second is the actual physical product you're selling: if the colour, texture or finish in the photo doesn't match exactly what the customer receives at home, the returns and negative reviews cost far more than the photo shoot you skipped at the start. The third is any piece where exact brand consistency (your precise corporate colour tone, typography, the framing you always use) is critical: there, a designer who knows your brand guidelines is still more reliable than a prompt, however good the model behind it is.
The approach that works best in practice isn't choosing between "AI or photographer", it's combining both with judgment: real photography for what genuinely represents your business (team, premises, flagship product, moments with clients), and generative AI for everything that's supporting material, variation and volume (backgrounds, mockups, ideation, secondary campaign pieces, tests before committing to a creative direction). That way you get the speed and low cost of AI without risking the trust that takes years to build and seconds to lose.
Frequently asked questions
Can I use AI-generated images on my social media without any issue?
For low-risk content, like a supporting image or an illustration of a concept, the actual risk is low, especially if you use tools with clear commercial licensing. For paid campaigns or wide-reach pieces, it's worth checking the tool's licence and avoiding images that directly imitate the style of a known artist or brand.
Which AI image tool is best for a small business?
It depends on the use. Adobe Firefly stands out for clear commercial licensing and integrates with Photoshop, which helps a lot with final retouching. Midjourney produces the most polished aesthetic finish for ideation and visual campaigns. DALL-E, built into ChatGPT, is the fastest for testing quick ideas with no learning curve. None of them replace an actual photo shoot of your product or your team.
Can people tell an image was AI-generated?
Less and less at a glance, but it usually shows in the details: text inside the image, repeated objects with symmetry errors, reflections that don't match the scene's lighting. Platforms like LinkedIn and Meta are also starting to add automatic "AI-generated content" labels whether you want them or not, so being upfront with your audience tends to be the safer bet in the long run.