Generate AI images on your device: Keep prompts local
Use a model on your device in the image workspace to generate from your prompt. Prepare the model files, make your first image, then reuse its settings for another attempt.
First, open /#/image-generation/ to start in the image workspace.
1. Open the image workspace
The link above opens Image generation directly. You can also open More Actions in the sidebar and choose Image generation. The workspace opens on a session, with settings on the left and execution, previews and results on the right. This is separate from generating an image inside an Ollama chat.
2. Prepare and select an image model
Open Components, then check that Compute with shows This device; switch it back if an earlier session restored another destination. Then choose Prepare a model beside Image model. In Models, choose an available model or download the files for a catalog entry to a supported destination, then select it for the workspace. Local-only cannot download catalog model files from external hosts; import a model folder obtained separately in Models. Check the model’s file size and required components before downloading. Catalog entries do not guarantee that every device or input will work.
3. Enter a prompt and generate one image
Leave Images per run at 1, enter a Prompt, and choose Generate. The Execution, previews and temporary results panel shows progress, then a preview. The result also appears under Runs and Gallery when saving is enabled and the session can save it. If Generate without saving in this workspace appears, choose it only for a temporary result and download any image you want to keep. You can change the prompt during a run; those edits apply to the next generation.
4. Inspect the result and try another version
Open the result card to inspect its prompt and settings. Choose Reuse prompt only to bring only its text back to the editor, or Reuse saved settings for the full saved setup. Use the viewer’s download menu to save an image as a regular file. Export session is a separate ZIP action for the session.
Keep the work you want to return to
The session draft and results need browser storage (OPFS) and Web Locks. When session storage is unavailable, the workspace may offer Generate without saving in this workspace; that result is not a saved session. Image sessions use an experimental data format that may change without migration. Save important images, prompts and settings outside Naidan. Session ZIP exports include inputs and outputs and local model paths, but not the model weights; a regular chat backup does not preserve the whole image session.
After your first image
The same workspace can generate several images, search a gallery, mark favorites, compare two saved images, and reuse an image as an initial or reference image. An initial image and a reference image serve different purposes and are optional for a first text prompt.