Brand and products
Using a product
How a product rides your messages, and what it tells Nomi.
Last updated 2 October 2026
Picking a product is how Nomi knows what the ad is for. It is the one piece of context you attach by hand, and it is worth doing before the first sentence.
Pick one from the composer
Open the picker
On New task, press Product right of the paperclip. Once one is picked the pill shows its face and name; with several it shows their faces and how many. Inside a session the field is one line, and the product is the paperclip's third row, with the product's face and name: press it to open the same list.
Search for it
The list belongs to the active brand, with a search field above it. Typing searches the workspace rather than the first screenful, so a long catalogue still finds its tail. Anything already picked stays in the list, under In this session.
Pick as many as apply
A duo or a three-product bundle is ordinary. Eight on one message is the ceiling.
To take one back off, open the pill and press its row again; nothing asks you to confirm.
Typing @ never offers a product. It names pictures already on the message. The Product pill is the only door.
What the product sends
Every message you send while the pill is there carries the product's name, whether it is a physical product or software, and its page address if it has one. Nomi is told this is context rather than a picture, and that it may read that page for detail you have not mentioned: the claims, the price, the tone.
Changing the selection applies from your next message forward. A message already sent keeps what it was sent with, so taking a product off never rewrites the conversation behind you.
Choose the pictures that matter
Selecting a product provides context and makes its photographs available. It does not automatically attach every photograph to your message. Name the pictures you want to use, or let Nomi select the relevant product photographs while preparing a generation in a conversation.
If a required product picture is missing, has unreadable dimensions or is smaller than 400 pixels on its shorter side, the generation stops before the image model is called. Replace it with a usable photograph and try again. A short side of at least 1,000 pixels gives the model more detail to work from, but does not guarantee accurate fine print.
Naming one picture
When a product has pictures, they sit as a small stack beside the pill. Press one and its name is written into what you are typing, along the lines of @acme-box. A lit card means that word is in your prompt, and pressing it again takes it out.
When it has no pictures
Type @ with that product picked and the list says so, and offers to fetch one from its landing page. If there is no page, or nothing usable on it, it says that instead.
Still missing something?