Ask for a campaign, not for an image.
It checks what it knows, reads what your market is running, and comes back with the argument before it draws anything.
You do not need another picture. You need to know what to run
Ask a generator for an ad and you get an ad. That is the easy half, and it has been the easy half for two years now. The hard half is the sentence before it: which angle, aimed at whom, against what everybody else in your category is already saying.
Most teams solve that in a meeting and then spend the week executing a decision nobody wrote down. Nomi does it the other way round. You say what you are launching, and the first thing that happens is not a render. It is the reading: what it already knows about your brand, what the product actually is, and what your competitors have been running while you were busy.
Every step is on the page while it happens, with the sources beside it. Nothing goes behind a spinner and comes back as a finished opinion you have to take on faith.
| Asking for a campaign | A prompt box | The strategist |
|---|---|---|
| What comes back first | A picture | The argument, and what it is based on |
| What it read | Your sentence | Your brand, the product, and the ads running in your market |
| Where the angle comes from | The model's guess at your category | The gap between what your rivals run and what nobody has taken |
| Changing your mind | A new prompt, from nothing | A sentence in the same thread, and the work follows it |
| Seeing what it is doing | A progress bar | Every step, with its sources |
| What it costs to disagree | The whole render | Nothing, until you approve the card |
The plan arrives before the pictures
Six brands selling the same thing usually sell it the same way. The useful answer is rarely another execution of the angle everybody is already running, and a tool that has not read the category cannot tell you that, because to it every angle looks equally fresh.
So a session comes back with the argument first. Here is what the market is saying, here is where the gap is, here is what I would run and in what weight. You change it in the same thread, in a sentence, and the work follows the change rather than starting again.
Say what you are launching
The way you would brief somebody: what the product is, who it is for, which market, what is happening on Thursday.
Watch it read
It checks what it holds about your brand, opens the product, and reads what is running in your category. The steps and their sources are on the page as they happen.
Read the plan
The angle, the reason it is open, and the campaign in the weight it recommends. Argue with it in a sentence.
Approve what gets made
When it decides something is worth drawing, a card appears with the idea, the angle and the hook, and the prompt in a field you can edit. The button names the price. Nothing is spent until you press it.
Change your mind while it is still working
You do not have to wait for a reply to finish before adding something. Send a correction mid-reply and it goes into that reply rather than starting a second one, so it lands while the work is happening instead of after it, which is the difference between steering and starting over.
Hover your own message and it says which of two things happened: read mid-reply, or not read at all. The second is normal on a short answer with no gap to look in, and it says so rather than letting you assume you were heard. Five corrections to one reply is the ceiling, because a whole conversation pushed into one reply has stopped being a correction.
Stop ends the reply on your screen. Nomi finishes what it had already begun and keeps what it had written, so stopping is not an undo.
When the argument is settled, skip the conversation
Beside the field are two lanes. Agent sends your sentence to Nomi for thinking. Image renders it straight away, with the shape and the engine your call rather than Nomi's, and the grid taking the screen while the field docks over it.
Use the first when you want an angle you have not had. Use the second when you know exactly what the picture is and you want four more shapes of it before lunch.
It keeps the thread, and then it keeps the brand
A session is one job with one brief. What you correct inside it holds for the rest of it. What you correct about your brand is written into what Nomi believes about you, under the subject it is about, and it holds for every session after it.
That memory is a page you can read. Open Brand, then Memory, and every belief is there in the words it was learned from, with the day it was learned. Anything Nomi wrote itself, or read off a web page, is refused before it can be stored, so what is in there came from you.
Start a second campaign as a second session rather than continuing the first. One session is one brief, and that is what stops last month's angle leaking into this month's launch.
What it will not pretend
It reads what is running, not what performed. Nobody outside Meta can see a competitor's results, and a tool that implies otherwise is guessing at your expense.
It works from what it has been told. A fact that is in your head and not in the brief is one it asks you for rather than invents.
It plans and it makes. It does not buy the media, and it does not report on what the media did.
What you would tell a colleague. What you are launching, who it is for, which market, and when. It asks for what it is missing rather than filling the gap itself.
It reads before it writes: what it holds about your brand, the product, and what your competitors are running. The plan comes back before anything is drawn.
Yes. Every step is on the page while it happens, with the sources beside it.
Say so in the same thread. The work follows the change rather than starting again, and disagreeing costs nothing because nothing has been rendered yet.
Yes, and your message goes into the running reply rather than starting a second one. It tells you afterwards whether it was actually read.
Not in this lane. When it decides something is worth drawing, a card appears with the prompt and the price, and you approve it. Declining leaves a line in the conversation saying it asked and you said no.
It remembers what you taught it about your brand, not the brief. A second campaign is a second session, so neither brief drifts into the other.
The Library, the moment each job exists. Sessions are conversations, not folders.