Copy & Use Prompts
Use gallery prompts in LocalBanana, check missing references and settings, and compare one change at a time. Copying text alone cannot recreate an image.
Find an image and inspect its prompt
Open a gallery image to read the displayed prompt and model label. Use prompt opens a new generation draft inside LocalBanana; Copy puts only the displayed text on your clipboard. Neither action is a complete record of how the example was made.
Generate from a gallery example in LocalBanana
- Open an image you want to learn from and choose Use prompt. The generation panel opens on the same page.
- Read the prefilled text. Check the model and aspect ratio: available, recognized values can be carried over, but missing values need your choice.
- If the prompt mentions an uploaded person, product, pose, or style image, attach the required reference yourself with Add image. The original reference files are not included with Use prompt.
- Review the selected model’s available size and quality options, reference attachments, and displayed cost or account requirements before generating.
- Keep the first result for comparison. Check the requested subject, framing, lighting, and any text before changing one part of the prompt.
Why the same prompt can produce a different image
A useful example pairs an output with enough information to explain it. A prompt alone may omit the original inputs or settings. Treat unknown settings as unknown rather than guessing that the displayed image used today’s defaults.
- Prompt and language: save the exact text used in your own run. Translating or shortening it changes the input.
- Model and version: record what you selected. A model name in an older gallery item does not guarantee the same version is still available.
- Reference images: save the files and check which ones are attached and in use. A described reference is not an attached reference.
- Aspect ratio, size, and quality: record the options actually exposed by that model. Use prompt does not copy every original setting.
- Output and date: keep the generated file alongside your notes. Randomness and provider updates mean that matching visible inputs still does not guarantee the same picture.
Make one change you can judge
For a first variation, keep the subject, model, references, and settings fixed and change one phrase—for example, soft window daylight to warm side lighting. Compare the face, framing, and shadows as well as the desired lighting change. This is a suggested exercise, not a tested result.
If preserving a particular face or product matters, use the original image as a reference in a model that accepts images. Reusing its prompt does not preserve its identity. Keep the source file so that later edits can be checked against it.
Worked examples with actual outputs
These September 3, 2026 experiments publish their inputs, settings, first accepted outputs, and limitations. Use them to understand the method; a fresh run is a new experiment.
- Same prompt, three image generators — An original portrait, a text poster, and a character sheet show why matching the input does not make different models produce the same image.
- Background replacement and identity preservation — One portrait reference is edited across three routes. Compare what changed in the person as well as the requested background.
Using the text in another tool
You can copy the prompt into another generator, including Gemini or AI Studio where available. Check that tool’s supported inputs and parameters, supply your own references, and review its account and cost requirements. Model-specific flags and output quality do not transfer automatically.