Generate an image
Open in AgentumPurpose
Parisa wants a first poster idea for the autumn sale in minutes, not hours. Create or edit pictures by describing them, starting from a ready-made prompt template, a quick sketch or your own words.
Before you start
Choose the workspace you want to work in. Image creation needs an image model that your plan allows. Settings, "Models" shows the default image model under "Image generation"; "Automatic" follows the model you chat with.
Steps
- Open "Images" in the sidebar. At the top there is a message box, a search box ("Search templates"), a "New" button, a "Draw" button, a row of categories ("Image template categories", for example "Trending", "Sketch", "Anime", "Product photo") and a grid of template pictures.
- Pick a category, or type in "Search templates". Clearing the search brings the full category back. More templates load as you scroll.
- Select a template. Its prompt is copied into the message box above so you can read and change it. Nothing is generated yet.
- Edit the prompt if you wish, and attach a picture with "Add to message" if you want it edited or used as a reference. To sketch your own starting point instead, select "Draw" next to "New". The "Drawing canvas" offers "Pencil", "Text", "Shapes" (line, arrow, rectangle, ellipse, triangle, diamond, star and heart), "Eraser", "Move text" and "Pan canvas", plus "Undo", "Redo", colors, sizes, zoom, "Upload image" and "Clear canvas". Select "Add to message box" and the drawing becomes an image attached to your message. It stays in your browser until you add it. Select "Send message". A new chat opens and the agent creates the picture. You can also skip templates and describe the picture straight in any chat, for example "Draw a poster for our autumn washing machine sale".
- In the chat, the picture first shows as "Creating image", then as "Image created". Select the picture to see it larger, or the download button in its corner ("Download") to save it. Select "Image created" to open its activity details, where "Model: [name]" says which image model made it. One request can produce up to four pictures. To change the result, reply with what to adjust and the agent makes a new version.
- To save your own template, select "New". In "New image template" fill in "Name" (up to 120 characters) and "Prompt" (up to 12,000 characters). "Preview image URL (optional)" takes the complete HTTP or HTTPS web address of a sample picture; it is only the thumbnail of your template and does not generate anything. Select "Create template". The template is saved for your workspace and the page opens its category. "Cancel" closes the dialog without saving.
Expected result
Sending opens a chat where the agent creates the image and shows it in the conversation. The prompt of a template you picked is only a starting point, and a template's preview picture is not the result of your request. Saved templates stay in your workspace for later use.
Troubleshooting
| Problem | What to do |
|---|---|
| "No templates match your search." | Use fewer words or clear the search. |
| "No image templates in this category yet." | Pick another category. |
| The agent answers with text and does not draw | Choose a chat model with the "Tools" badge and explicitly ask it to generate the image. If it still cannot, check the image model in Settings, "Models", and the agent's configuration. |
| "The media service is busy. Please try again shortly." | Wait a moment and send the request again. |
| "Enter a name and a prompt. The preview must be a web address." | Fill in both fields, and make sure the optional preview address is a complete HTTP or HTTPS web address. |
| "The selected model cannot read images. Add a text prompt or choose a model that supports images." | You attached a picture to a model that cannot read pictures. Pick a model with the "Vision" badge. |
| "Your current plan does not include this feature.", "You have reached your usage limit. Try again after it resets." or "You have reached your usage limit. It resets at [time]." | See Understand plans and usage limits. |
| The agent does not draw in a private workspace, project or chat | Image generation needs a private image model while Data Privacy Mode applies. See Use Data Privacy Mode. |