You bring the model
Image generation runs on a model you have added yourself, the same way every other model in nemu works. Add one in the console under Models, with its type set to Image, and it becomes available in chat straight away. If you have no image model, the assistant says so and stops rather than guessing. Nothing is generated and nothing is charged.Choosing which model runs
The first time a chat needs an image, you are asked which of your image models to use. The prompt has one entry per model, and each model offers two ways in:
Pinning is remembered per chat, never across your whole account, so a new chat
asks again. Leaving the page while the question is open counts as a denial: the
request is dropped rather than left waiting.
The gallery
Every generated image is listed at chat.nemu.cc/gallery, newest first, with the prompt that made it, its size and the chat it came from. Select any number of images to delete them and free the space. Images are not deleted on a timer. They stay until you remove them, which is why the gallery has a storage meter in its header.Storage limits
Each plan has an image storage allowance, listed in Plans, along with a daily and weekly cap on how many images you can generate. Storage is checked before anything is generated. When you are out of room the assistant tells you to clear some images and links you to the gallery, and no request is sent to the model, so a full gallery never costs you anything.Images in a Computer project
When a chat has a Computer project open, an image can be written into the project as a real asset file rather than only shown in the conversation. Ask for it in the project, for example a hero image for a page you are building, and it lands in the project tree ready to reference from your code. An image saved into a project is a copy. Deleting it from the gallery does not remove it from the project.Custom agents
generate_image is available to custom agents as a tool. An
agent can only use a model you have already pinned in that chat, because an
agent running in the background has no way to ask you which model to use.