Sessions
Agent and Image
A session has two controls that look alike and do different things: a switch at the top that decides what you are looking at, and a strip beside the field that decides where your message goes. This page is mostly about the second one.
Last updated 20 August 2026
The switch at the top is not a lane
Above the conversation sits Strategise and Generate. It changes the layout of the session and nothing else. Strategise gives the conversation the whole width. Generate splits the screen: everything the session has made on the left, the conversation in a rail on the right, with a divider you can drag. The setting belongs to the session and is remembered, and the first render in a session moves it to Generate on its own.
The strip beside the field is the lane
At the left of the composer's control row are two tabs, Agent and Image. The live one shows its name, the other is just its icon. This is where your message goes when you press Enter, and the two go to genuinely different places.
Agent
Enter sends the sentence to Nomi. You get an answer, and when it decides something is worth rendering it does not simply spend: a card appears on the composer with the idea it is testing, a few words for the angle and the hook, and the prompt in a field you can edit first. You choose the engine there, and if Nomi proposed several you choose whether to render this one or all of them. The button names the price. Decline leaves a line saying it asked and you said no.
Image
Enter renders, straight away. There is no reply to read, so the conversation slides out of the way and the grid takes the screen with the field docked over it. Two extra controls appear beside the tabs: the shape it renders at, and the engine that draws it. Both are your call here, where in the Agent lane the shape is Nomi's and the engine is on the card.
In the Image lane, Enter spends credits. Nomi asks first only if you have set a price above which it should check, and never if the switch under the field is set to Bypass permissions.
Your prompt does not appear in the conversation in this lane. The creative is the record, and it carries the prompt as its title. Products and pictures you attach still travel with the render.
Which one
Use Agent when you want thinking: an angle you have not had, several takes on one product. Use Image when you already know the picture you want.
The lane strip is not drawn on a phone. There is no room for the grid beside it, so the conversation gets the whole screen and renders go through Nomi's card.