Using Gallery References

Choose a Reference for a Specific Reason

Open a gallery image and compare the visible result with its prompt. Before reusing it, identify three things: Browse the gallery.

• Subject: What should the image contain, and does your task require a particular real person or product?
• Visual direction: Which composition, lighting, palette, or medium do you want to borrow?
• Must-keep details: Which face, pose, garment, label, or exact words would make the result unusable if changed?

Use the Prompt in LocalBanana

Once the example fits your task:

1. Open its detail view and read the full prompt, including any instructions that refer to an uploaded image.
2. Choose “Use prompt” to open the image composer on the same page.
3. Check the supplied prompt, selected model, and aspect ratio. Replace example-specific details with your own requirements.
4. Add any necessary source images yourself, check that the model supports them, then review the cost and generate.

This action reuses text and supported metadata; it does not attach the original reference images or restore all settings. If a prompt says “use the uploaded face” and no source image is attached, you do not yet have the required inputs.

Change One Thing and Check the Result

Keep the parts you want to compare stable. Useful changes include:

• Subject: replace the example object with your own. For an identifiable product, supply a source image instead of relying on its name alone.
• Style: replace the rendering medium, such as oil painting with watercolor, while leaving the composition unchanged.
• Editing: state the requested change and what must stay the same. For a background edit, check the face, pose, clothes, hair edges, and shadows against the source.
• Format: choose the aspect ratio in the composer controls and adjust the framing instructions. Merely writing “16:9” in a prompt does not set the output dimensions.

Why the Same Prompt Can Produce a Different Image

Prompt text is only part of the recipe. Model and version, reference images, aspect ratio, available settings, and generation variation affect the output. When references or settings are missing, treat the example as a visual direction, not an exact reproduction target. A different model may also support a different kind of editing.

• Copy and adapt prompts — Use the input checklist, or copy the text for an external tool and configure its inputs separately.
• Same prompt, three AI image generators — Compare actual outputs to see how models interpret the same words differently.
• AI image editing prompts, tested — Inspect source images and edited results before choosing a preservation strategy.