I Do Almost All of This on a Phone Now

For the first month I did this at a desk, because that is what generating images looked like in my head: a big screen, a proper keyboard, a grid of results. Four months later I would guess nine in ten of my generations start on a phone, standing up, usually somewhere I would otherwise be scrolling.

I did not expect that, and the reasons turned out to be more interesting than “convenient”.

The thing that actually changed: it runs in the browser

There is no app. That sounds like a shortcoming until you notice what it means in practice: nothing to install, nothing in a purchase history, nothing sitting on a home screen with a recognisable icon, and the same account you already have. For a category people mostly use privately, that is not a minor detail. I cover the rest of that in the privacy write-up.

The practical consequence: you add it to the home screen yourself if you want it there, and it opens like a page rather than announcing itself.

Why a phone turns out to suit this

The real reason is timing rather than hardware. Ideas for images do not arrive at a desk. They arrive in a queue, on a walk, mid-conversation, and the gap between having one and being at a computer is where most of them died.

There is a second, smaller effect I noticed after a few weeks. A generation takes long enough that at a desk I sat and watched it, which is a strange thing to do and makes each one feel expensive. On a phone I lock the screen and do something else, because the work happens on the server rather than on the device. Same wait, completely different feeling about it.

Three habits that made it work

The first month on a phone was worse than the desk, and it was mostly self-inflicted. Three things fixed it.

Dictate the prompt. This is the big one. A thumb keyboard quietly pushes you towards eight-word prompts, and eight-word prompts get you generic results. Speaking one gets you thirty words without noticing, and thirty words is roughly where the description stops being a label and starts being a scene. It also produces a different register: spoken prompts read like someone describing a photograph, which is closer to what works than a tidy written list of nouns.

Keep a note file of prompts that worked. Retyping a long prompt on a phone to change one word is unpleasant enough that you will not bother, which means you stop iterating, which is the whole skill. Paste from a note instead and the friction disappears. This is the smaller cousin of keeping an actual prompt library, and on a phone it matters more, not less.

Stop trying to judge results at arm’s length. A phone screen flatters an image. Things I was happy with on the phone and then opened on a laptop had visible problems: hands, background detail, an off-looking texture. If the result is going anywhere real, look at it once on a bigger screen before deciding it is finished.

What is genuinely worse

Comparing options. Four results side by side is how you actually develop taste, and a small screen makes it a scrolling exercise instead of a glance. When I am trying to choose a direction rather than produce a final thing, the desk still wins.

Long, precise prompts with several clauses are also harder to edit, even dictated, because moving the cursor into the middle of a paragraph on a touchscreen is its own small ordeal.

And video is a mixed case. Setting up a clip is fine on a phone, but reviewing one properly means watching it a few times at a decent size, and a clip with sound especially deserves better than a phone speaker in a noisy room.

The vertical thing nobody mentions

Worth its own paragraph because it took me weeks to internalise: if the image is going to be looked at on a phone, generate it in a vertical aspect ratio from the start. Cropping a landscape image to a portrait shape later loses the composition, and the generator will happily give you a 9:16 frame that was composed as one.

Sounds obvious written down. I made maybe sixty landscape images before it occurred to me that I never once viewed any of them in landscape.

Costs are identical, in case you were wondering

Same generator, same models, same coins: a Standard image is 40 coins whichever screen you started it on, and there is no mobile surcharge or app-store cut, because there is no app store. The full breakdown is in the costs piece if you want the arithmetic rather than my word for it.

If you have only ever done this at a desk, the experiment worth running is one week of dictating prompts on a phone while walking. My prompts got longer, more specific and noticeably less like a shopping list, and I do not think I would have got there by typing. Start in Fellowi Images in your mobile browser, and the prompt examples are a reasonable thing to read aloud to yourself first.