You can schedule illustration work on your own computer instead of repeating an image prompt for every publishing session. In OGAD (Off Grid AI Desktop), a God routine can request image generation and save the resulting images in its conversation.
This needs Pro for the routine, a suitable chat model for the instruction, and an available image-generation model. Keep the computer awake and OGAD running. A text reply that says “images created” is not enough: check the actual generated image and its saved result.
Download OGAD | Desktop beta 122 release notes
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@getoffgridai.co, join our Slack community, or talk to us on Reddit.
Start with one illustration, not a whole campaign
Choose a single article and a clear visual purpose. An illustration might explain the setting, introduce a character, or provide a cover composition. It should not pretend to be a screenshot of a feature that does not exist.
Write a short brief with the subject, setting, composition, palette, and things to exclude. If the article needs readable labels, add those through a suitable layout process after generation instead of relying on the image model to spell them correctly.
For example: “Create one editorial illustration of a small studio planning a project at a shared table. Use soft daylight and restrained green accents. Leave clear space at the top for a heading. No logos, watermarks, readable text, or claims about a real company.”
Check the local image setup
Install OGAD and confirm that ordinary image generation works before adding a schedule. Download the image model and any required components while you have internet.
Choose the intended local image engine and its defaults in the app. The routine's image request uses the configured generation path. Do not assume that every custom width, seed, or model setting written in a prose prompt will be passed through as a supported generation option.
Generate one manual test image with the same visual brief. Check that the computer has enough available memory and storage for that model. A working text-chat model does not establish that an image model is installed or ready.
Create the illustration routine
Open God settings, expand Scheduled tasks, and choose Add a scheduled task. Give it a name you can recognize, such as “Friday studio article illustration.”
In What to do, request an actual image and describe the deliverable:
Generate one image for the studio-planning article using this brief: [insert the visual brief]. Use the available image-generation tool. Save the generated result in this conversation. Report a failure if generation does not complete. Do not publish the image or claim a finished file exists unless the generation result confirms it.
Choose When. A one-time schedule fits a specific article; selected weekdays fit a recurring editorial slot. Choose Add and leave the task enabled.
Keep the brief self-contained. A vague reference to “the article from earlier” may pull in the wrong conversation or omit the details the image needs.
Use Run now and inspect both stages
Choose Run now for the first check. The routine first processes the instruction, then starts the requested image job. Image generation may wait for other local work, so the text response and the image result do not necessarily arrive together.
Open the conversation and find the actual generated image. Check the job state and any error text. If the job failed or was cancelled, the correct outcome is a failed attempt, even if an earlier model sentence said the image would be ready.
Save or open the result through the app's image controls. Confirm that the file is the expected image and that it is usable at the size your article needs. Do not build a publishing process around a claimed count alone.
Review the illustration as editorial content
Check the subject, composition, and unwanted details. Look for distorted objects, accidental text, misleading symbols, and anything that conflicts with the article. A visually attractive result can still explain the wrong thing.
Write alt text for what the image actually shows. Label a conceptual illustration as such when readers could confuse it with a product screen or real event. Do not add a numerical performance claim to the caption unless you have separate evidence for it.
For a series, keep the useful parts of the brief consistent. Then review every result. A repeated prompt does not guarantee identical characters, layouts, or style across separate generations.
Fix failed or missing output
| Symptom | Next check |
|---|---|
| Only a text response appears | Image request and job result in the conversation |
| Generation cannot start | Installed image model and selected engine |
| Job waits for a long time | Other local model work and available memory |
| Routine never runs | Enabled state, schedule, app, and awake computer |
| Image is saved but unsuitable | Revise the brief and generate a new version |
Turning off or deleting a routine before its queued image work starts can prevent that work from being made. Check the recorded state instead of assuming every queued request must eventually finish.
Keep publishing as a separate step
The routine creates illustration work in OGAD. It does not, by itself, upload a finished image to your blog or approve the article around it. Review the image, caption, and article placement before you publish.
Set up one illustration routine in OGAD, test it now, and verify its saved image. Once that works, schedule the next useful visual with an equally specific brief.
Top comments (0)