What I Actually Use This For, Which Is Mostly Boring

I went back through a month of generations expecting to find pictures. Mostly I found utilities. Almost nothing I have made was intended for anyone to look at as an image, and the cases that turned out genuinely useful are the ones nobody writes enthusiastic posts about.

So, an honest inventory.

Showing someone the thing in my head

This is the biggest category by a distance and it is barely about pictures at all.

Describing a visual idea in words is slow and unreliable. Two people agree on a sentence and picture different things, and you discover the mismatch three weeks later. Generating a rough version takes forty seconds and forty coins and collapses the whole misunderstanding immediately, even when the image is mediocre. Especially when the image is mediocre, actually, because nobody mistakes it for a proposal.

Concrete cases from the last month: what a room would look like with a different wall colour, roughly what a piece of furniture I was describing to a joiner looked like, and a layout idea I could not explain over a call. None of those images survived the conversation they were made for.

References for someone who can actually draw

A friend who illustrates asked me to stop sending her descriptions and start sending her rough generations instead. Not to copy, which would be pointless, but because a bad picture of the right pose is a more useful brief than a good paragraph.

This is the use that has aged best for me. It is also the one that makes the “is this replacing artists” question more complicated than either side likes, because in that exchange the generation is doing the job a stock-photo search used to do badly.

Backgrounds and filler that would otherwise be stock

Slide backgrounds. A texture behind a title. Something to fill a space in a layout that would otherwise be a grey box or, worse, a stock photograph of people in a meeting laughing at a laptop.

Unglamorous, and the honest reason it is worth doing is not that generated art is better than stock. It is that it can be specific to the thing, and it takes a minute instead of twenty minutes of scrolling.

Thinking, when I do not know what I want yet

This one surprised me. Sometimes I generate four versions of something not to use any of them but to find out which direction I actually prefer. Seeing four options is a faster way to discover your own taste than staring at a blank page trying to introspect.

It costs 160 coins to have an opinion you did not have before, which is a strange sentence to write and has genuinely been worth it more than once.

The private ones

A reference for a tattoo I have not booked. What a haircut might look like. A picture of a place from a story I was writing that nobody else will read.

Worth saying plainly because these get left out of every article about this: a large share of real use is private, mildly self-indulgent and never shown to anyone. That is a completely legitimate way to use the thing, and it is why the privacy mechanics matter more than the feature list.

What it is genuinely bad at

Worth being specific, because the failures are predictable rather than random.

• Anything with text in it. Signage, labels, a poster with words. It will produce letter-shaped marks and you will not want them.
• Diagrams and anything that has to be correct. It generates plausible, not accurate. A diagram that looks right and is wrong is worse than no diagram.
• A specific real product. Not the category, the actual object. It will give you a convincing thing that is not that thing.
• The same face twice, unless you carry a reference image, which is a whole discipline of its own covered in the consistency post.

The pattern in that list: it is bad wherever plausible-but-wrong is worse than absent. Keep it away from anything factual and it stops disappointing you.

The part about work

Worth being exact here, because I had the wrong idea at first. Our own Terms place no restriction on commercial use, so paid client work is fine as far as Fellowi is concerned. What is not automatic is everything outside Fellowi: a platform or a contract may add rules of its own, and whether you hold copyright in a machine-made image is genuinely unsettled. Finding that out afterwards is the unpleasant version.

And the disclosure question has a simple test, which is whether anyone would feel deceived on finding out. A slide background, no. An illustration presented as commissioned artwork, yes. I wrote about where that line sits in the piece on whether this is cheating.

None of this is the future-of-creativity story. It is a tool that made it cheap to show instead of describe, and most of the value has turned out to be in conversations rather than in pictures. If you want to try the boring version first, a mock-up of something you were about to explain in a paragraph is 40 coins in Fellowi Images, and the first-week account covers the rest.