- ProblemLogo, product photos and posts: what comes out usable and what gives it away.
I set out to produce my business's entire look with artificial intelligence in one afternoon, without commissioning anything. The social media posts came out well, the service photos so-so, and the logo I ended up redrawing with my own hands. This is what I learnt along the way.
The whole-brand-in-an-afternoon promise
"Create your brand's entire identity with AI in one afternoon and save on the designer." You'll have seen it in different words with the same music: logo, colours, photos and social media templates in a couple of hours.
The business I tested it on is the one from the series: Mostrador Digital, a service that builds FAQ assistants for local shops, whose launch I describe in Getting the business off the ground. I needed a logo, images showing the service and posts for social media, and I set myself a rule: the machine produces everything visual; I only ask, choose and touch up.
I gave myself one afternoon, about 4 hours. It ended up being about 12 hours spread over several weeks: the logo in the second week of the launch, the photos in the third and the posts once I was getting the word out. The afternoon stayed in the video title.
What a business needs visually
For a small business, the essentials are few, but each has its demands.
- A logo that can be edited, that looks good small (in the browser tab icon, in a social profile picture, in the corner of the chat) and that doesn't resemble anything that already exists.
- Product or service photos that show what the customer will get, no more and no less.
- Social media posts, which get renewed every week and where volume matters.
My criterion for calling something "usable" is simple: I'd put it on my business without embarrassment, in front of someone who's going to pay me.
How images are generated and which tools I tried
For beginners, the idea is simple: generating images with artificial intelligence means describing in words what you want (that's the prompt) and the tool turns it into an image in a few seconds. You can also start from a photo of your own and ask for changes to it. Under the hood, the model has learnt from millions of captioned images which shapes and colours go with which words, and from there it "paints" something new.
I tried three routes: a general-purpose assistant, ChatGPT, which generates images right inside the chat; a paid specialised tool, subscribed for one month; and an open model running locally, on one of my servers, because I already had the hardware and wanted to see if it held up in the comparison. I chose those three because they cover the typical cases: what you already have to hand, what you pay for quality and what runs at home without depending on anyone.
The logo
I started with the hardest part without knowing it was. My first request was something like: "Logo for Mostrador Digital, a service that puts chat assistants in neighbourhood shops. A speech bubble shaped like a shop awning. Flat style, two colours, white background".
Pretty things came out, and that's where the typical problems began. The first, the lettering: in half the versions the name appeared as "Mostardor Digtal" or with made-up letters. The second, that the result is an image, not a logo: there's no way to change a colour or separate the text from the drawing without redoing it, and to use it properly you need a vector file, which is a drawing made of shapes that can be scaled without losing quality. The third, size: the awning stripes, so charming when large, turned into a smudge at 32 pixels.
It took 23 rounds and about 5 hours. In the end I chose the simplest version from the paid tool and redrew it myself in a vector drawing program, cutting the awning down to three stripes and setting the name in an ordinary typeface.
Would I use it? I'm using it, but the final logo isn't the machine's any more: it's mine, with one of its ideas. And that, as we'll see in the legal part, is no minor detail.
The product photos
My product is a service, so the "product photos" were images showing what it looks like: a phone with a shop's chat open, a sign on the counter with the code to scan. I tried two things.
First, images generated from scratch: "a hand holding a phone in front of a bakery counter, on the screen a chat asking about opening hours, morning light". The atmosphere was lovely, but the text on the screen was hieroglyphics, the phone changed model in every image and in one of them the chat showed a "pay now" button my service doesn't have.
Then, retouched real photos. I took photos with my phone at the pilot bakery (with my friend's permission) and real screenshots of the assistant, and asked the tool to improve the lighting and place the screenshot inside the phone screen. This worked a lot better, with a catch that cost me an extra round: in one of them the tool also "improved" the screenshot and swapped the assistant's answer for a nicer one that wasn't the real one.
And that's the limit I set myself: the customer buys what they see. If the image shows something the service doesn't do, disappointment follows on its own. I kept the real photos retouched only for lighting, six in total, about 4 hours of work.
The social media posts
This is what turned out best. I prepared the text with Claude from my own material, which is the key to creating content with artificial intelligence that doesn't sound like anyone else's: the odd questions the bakery assistant received, what a hardware shop owner told me when I showed it to him, or the Sunday opening hours scare. The images, with the general-purpose assistant, working on the real photos.
I asked for different versions for each network, because you don't write the same way for Instagram as for LinkedIn. When using artificial intelligence on a business's social media, this is what makes the difference: the same message, told the way it's told in each place.
What didn't work: generic posts like "Did you know AI can transform your business?", which the assistant proposes by the bucketload and which nobody commented on. What worked best: the real screenshot of a bakery customer asking whether the dog could come in, with the assistant's reply. Preparing ten posts took me about 3 hours, corrections included.
What gives away the machine
In my results, the signs that betrayed a generated image were always the same: lettering that almost says what it should, extra fingers holding the phone, a plasticky sheen on everything, shadows that don't match the light and background shelves that melt if you look closely.
I noticed because I knew where to look. What was interesting was what others noticed: I showed the images to my baker friend without saying anything and he immediately pointed at the two generated from scratch: "that's not my shop, is it?". He couldn't say why, but he saw it. And if someone who isn't looking sees it, a customer will.
What you can and can't use in a business
A warning up front: I'm not a lawyer. What follows is how I've looked at it in practical terms, to stay out of trouble. The legal references are Spanish, because that's where I live and work.
First, the question of who owns the copyright to what artificial intelligence creates. In general, what's generated without a noticeable human creative contribution isn't usually protected by copyright, because most laws, including Spain's, protect works created by people. Translation: if my logo were exactly as the machine produced it, I'd struggle to defend it as mine if someone copied it. Whether I can use it is another matter, and often I can.
Second, the risk of resembling something that already belongs to someone else: a well-known brand, an artist's recognisable style, a character. One of the logo versions looked suspiciously like that of a well-known messaging app, and I discarded it without a second thought.
Third, each tool's terms of use, which say what you can do with what it produces, including for commercial purposes. I reviewed those of the three tools I used on 12 June 2026; they're worth rereading, because they change. It's an area still being decided in courts and legislation, so I don't take anything as settled. For a logo that's going to be your brand, the sensible thing is to consult a professional before registering it.
Cost and time versus commissioning it, and what I'd leave to the machine
I didn't ask any designer for a quote, so I'm not going to invent a figure to make the comparison look nice. What I know is my own: about 12 hours and €22 for the paid tool, plus the time spent redrawing the logo.
My split after the test: I leave posts, drafts and trying out ideas to artificial intelligence. Anything that is the brand, like the final logo, and anything that shows what the customer is going to buy goes through my hands or those of someone who knows what they're doing.
Community alternatives 0
Nobody has suggested an alternative to this case yet. You could be the first.
Suggest an alternative