A campaign brief describes what an audience should understand. An image prompt needs something more concrete: a subject, a scene, and a clear visual idea.

OGAD (Off Grid AI Desktop) can help you turn a written brief into image concepts and generate a first visual with a local image model. Use a local text model to plan the directions, then review one generated image against the brief. After setup, both stages can run on your computer.

Download OGAD

Desktop releases

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@offgridmobileai.co, join our Slack community, or talk to us on Reddit.

For a small creative team, this is a way to make early directions visible before spending time on final production. The output is a concept to discuss, not proof that the campaign will perform or a substitute for brand review.

What should you extract from the brief first?

Identify the audience, the message, the desired response, and the constraints. Keep those facts separate from the visual style you might choose. A beautiful image can still communicate the wrong idea.

Suppose a training company wants to introduce a workshop that helps small teams make project handovers clearer. The image needs to communicate shared understanding and continuity, while leaving space for a headline added later.

A useful concept brief includes:

Element	Question to answer
Audience	Who should recognise the problem?
Message	What should the image communicate?
Subject	What visible object or scene could carry that message?
Format	Where will the image appear?
Constraints	What must be included or avoided?

Do not start by listing every visual feature the model can generate. Start with the point the image needs to make.

What do you need in OGAD?

Use a local text model for the brief analysis and a supported local image-generation model for the visual. Each stage needs its own appropriate model files. Download the app, models, and required runtimes while connected.

The free core desktop app includes image generation on supported Mac and Windows computers. 0.0.54-beta.108 also provides Linux beta packages. Available processing backends and memory requirements differ by platform and model.

Choose an image model that fits your computer rather than assuming the largest option is necessary for an early concept. Keep enough free storage for model files and enough available memory for generation.

Use a readable saved brief. TXT, Markdown, DOCX, or a text PDF works for the text stage; scanned material may need checked text first.

How do you create distinct concept directions?

Attach the brief to a chat with a local text model selected. Check the extracted text, then ask for a small set of directions that differ in the central idea, not just colour or style.

Propose three visual concepts for this campaign brief. For each, state the message, visible subject, composition, and reason it fits the audience. Keep the concepts meaningfully different. Identify any assumption that the brief does not establish. Do not invent product claims or audience results.

For the handover workshop, one direction might show a continuous path, another a shared plan, and another a clear transfer between two work areas. These are candidate visual interpretations. Choose the one that fits the campaign’s tone and purpose.

Ask what each concept could be misunderstood to mean. A visual about transferring work could accidentally suggest abandoning responsibility. That is useful to notice before generating many versions.

How do you turn one concept into a prompt?

Describe visible elements and their relationships. Keep the prompt focused enough that you can judge whether the model followed the main idea. Add technical or brand text later in a tool where you control it precisely.

A simple concept prompt might be:

A clean editorial illustration of two work areas connected by one continuous green path, warm white background, clear composition, generous empty space on the right, no text.

Review whether the subject communicates the intended message without an explanation. If the idea depends on a paragraph of labels, it may be too complicated for a single campaign image.

Avoid asking the image model to invent an accurate product interface, chart, or testimonial. Use verified screenshots or data-driven graphics for those jobs.

How do you generate the first image locally?

Download and select a supported image model in Models. Open Chat, then use + > Generate image. Enter the prompt and review the available Image options before starting.

The chat image controls support image generation as a separate action from ordinary text chat. If several local image models are available, choose the intended model in the image options.

Start with a simple composition and a modest output size suited to the selected model. A larger canvas can increase resource use without fixing an unclear concept.

When generation finishes, inspect the image and use the generated-image view’s Download control to save a candidate. Keep the prompt with the file so you can understand how the direction developed.

How should you review the concepts?

Judge the image against the brief before judging small visual details. Ask whether it communicates the right message, fits the placement, and leaves space for the copy or branding you need to add.

Review question	What to inspect
Message	Does the main idea read clearly?
Audience	Is the tone suitable for the intended reader?
Composition	Is the subject visible at the final size?
Accuracy	Does the image imply a false product capability or result?
Production	Can you add the required brand elements cleanly?

Show a small number of directions with a one-sentence rationale. An unlabelled pile of variations can make review harder because people compare details without knowing which message each concept is testing.

How do you improve a weak result?

Change one important thing at a time. If the composition is busy, reduce the number of objects. If the subject is unclear, make the main object more specific. If the style is wrong, keep the subject stable while changing the visual treatment.

Record the reason for each revision. This helps you distinguish a concept problem from an image-quality problem.

Problem	Next change
Too many competing elements	Keep one central subject
The image needs explanation	Simplify the visual idea
Generated text is wrong	Remove text and add it later in your editor
The model cannot load	Try a smaller compatible model or reduce resource demand
A polished image misses the brief	Return to the message before generating more variants

Make one direction concrete

Download OGAD, analyse one campaign brief, and choose one visual idea to generate. Review it against the message before turning it into final creative work.

Keep both text and image models local for this workflow after setup. Remote model choices and later sharing have separate data paths. The useful result is a clearer creative decision, supported by an image the team can actually discuss.