How to use Meta AI Imagine and keep useful image prompts

To use Meta AI Imagine, write a clear visual request, submit it if your account offers image creation, and refine the result one change at a time. Keep the exact prompt, any settings you actually used and a note explaining which result worked. fenbs can keep that record outside the conversation so you can carry on with the next version later.

Meta documents its assistant in the standalone app and on the web in its Meta AI app introduction↗ (opens in a new tab). Start with the place you already use. Keep the first request and its revisions together, and keep your own record of the versions you review.

Where should you start an image request?

Start in the Meta AI standalone app or its web conversation, the two places described in the app introduction linked above. That page does not give a complete Imagine button sequence, an image-specific eligibility list or exact settings labels. Check the help available to your account before preparing a large set of images; if image creation is not offered, keep the written request for later.

1. Write down the purpose of the picture: a presentation cover, a draft illustration or a visual idea to discuss.
2. Describe the subject, background, arrangement, lighting and style in a short prompt.
3. If your account offers image creation, submit that request through the available control. Keep a copy of the exact text you used.
4. Check the returned result against the purpose before asking for any revision. If no image is returned, record that outcome instead of treating it as a completed generation.

This page focuses on a visual request and its revisions. For research, writing and planning requests, use the separate guide to prompts for Meta AI. More reusable prompt methods are collected in the AI prompts hub.

What should a useful visual prompt include?

Begin with what someone should see, rather than a string of praise words. “A blue ceramic mug on a plain kitchen counter, seen from the side” gives a reviewer something concrete to check. “An amazing professional image” leaves the desired result open to interpretation.

Say what must stay fixed and what can vary. For a hypothetical presentation cover, you might need an empty area for a title, a calm background and one clear subject. You might be willing to try different lighting. Those constraints make later feedback more precise.

Create a draft illustration for a presentation cover.
Subject: one blue ceramic mug on a light kitchen counter.
Composition: mug on the right; empty space on the left for a title.
View: side view, with the handle visible.
Style: simple editorial illustration with soft daylight.
Avoid: lettering, logos and extra objects.
This is a concept to review, not a finished advertisement.

Treat this as wording to test, not a guarantee of a particular output. Keep the request short enough that you can identify which instruction an image followed or missed. Add detail when the review reveals a specific problem.

How do you refine an image without losing the direction?

Choose the most important mismatch first. If the subject is wrong, fix that before adjusting shadows. If the composition leaves no space for your title, ask for the subject to move before debating color. A revision needs a clear reason, even when the request is only one sentence.

1. Name the version you are reviewing in your own notes, such as Mug A.
2. State what you want to preserve: the subject, viewpoint or empty area.
3. Request one change, such as a simpler background. Use only revision options your account actually provides.
4. Compare the new result with the earlier one and record whether the requested improvement happened.

For example, try “Keep the side view and the mug on the right. Make the background plain.” If the next result changes the subject too, record that failure. Do not silently rewrite the original requirements to make the new picture seem successful.

Keep the winning wording and the reason it worked with your other saved work in fenbs. That gives the next assistant a useful starting point even when you choose a different image tool.

What should you check before using the result?

Review the image at its intended viewing size. A small preview can hide unreadable lettering, inconsistent objects or details that become distracting on a large slide. Ask a person to check the actual picture, rather than accepting a text description of what the picture supposedly contains.

• Purpose: does the picture communicate the idea you wanted?
• Composition: is there enough room for text or a crop you plan to add later?
• Details: are objects, edges and any visible text acceptable for the intended use?
• Approval: who needs to review this draft before it goes into shared work?
• Version: can you identify the exact result that was approved?

Use a fictional subject for a first test. Do not upload private material just to discover whether a control works. For a broader decision about the assistant itself, see ChatGPT versus Meta AI; for a picture involving your own likeness, follow the separate Imagine Me guide.

What can you do when the request does not work?

Describe what happened before trying a fix. A missing control, a written answer instead of an image and a picture that misses your instructions are different outcomes. The next test should address the outcome you actually saw, without assuming a hidden account or technical problem.

Why should you keep a record outside the conversation?

Meta documents app and web chat continuity and selective remembered details. Its personalization explanation↗ (opens in a new tab) does not promise that memory will preserve every prompt and revision verbatim. A useful creative record should therefore say which wording you chose and why.

Record only settings you actually saw and used. If no size, model or other generation setting was visible, write “not recorded” rather than inventing a value. Preserve the original prompt separately from later requests so a colleague can distinguish your starting instructions from your corrections.

A simple record can include the image purpose, exact prompt, revision request, observed result and next change to try. Keep the image in an approved file location and note its name. The review notes should remain useful even when somebody cannot reopen your original conversation.

Purpose: presentation cover illustration
Version: Mug B
Starting prompt: [paste the exact request]
Revision: keep the side view; make the background plain
Visible settings used: [copy labels and values, or write none]
Result checked: mug on right; space for title on left
Image file: [name and approved location]
Next step: check the crop in the presentation

How to keep image prompts and progress with fenbs

Keep the work so you can carry on from exactly where you left off, on any device and in any other AI. Save the instructions that helped and the next small change to try. There is no need to copy every unsuccessful message.

1. Create a free account.
2. Save the image purpose and the prompt you want to use again. Give it a clear name, such as Presentation cover mug.
3. Add what worked, what still needs fixing and the name of the image file you reviewed.
4. Choose an AI from Integrations. Ask it to read your saved work before helping with the next version.
5. Before you stop, save the next step so you can pick up the work later.

fenbs does not generate images or import your Meta AI chats. Meta’s connector program↗ (opens in a new tab) is a limited developer preview, which does not establish a general Meta AI connection to fenbs. If you only need one picture and will not reuse the process, a short local note may be enough.

When an image belongs to a wider campaign, the guide to AI for Meta marketing covers the broader planning work. Keep this image record focused on the visual request and the evidence from your own test.