I’m leony, an indie maker who likes turning repeated experiments into small, useful tools. One problem I kept running into with Nano Banana image generation was the blank prompt: I knew the kind of image I wanted, but not how to describe the subject, lighting, composition, and constraints in a way that gave me a useful starting point.
So I built the Soutine Nano Banana Prompt Library, a collection of more than 1,400 prompts with real generated previews. The goal is not to present one “magic prompt.” It is to make the relationship between exact wording and a visual result easier to inspect.
Why show the result beside the prompt?
A prompt without an example is difficult to evaluate. A generated image without the prompt is difficult to reproduce. Putting them together makes the library more practical: you can look at the composition, notice the language used for the subject or materials, and then copy the complete prompt as a baseline for your own experiment.
The collection is organized around common creative tasks rather than a single style. You’ll find examples for portraits, product shots, image restoration, anime, interior spaces, and surreal edits. That range matters because each category tends to need different kinds of detail. A portrait may depend on expression and lighting; a product shot may need surface descriptions and a controlled background; a restoration prompt may focus on preserving identity while repairing damage.
The browse copy → try workflow
My preferred workflow is simple:
- Browse a category that is close to the image I want.
- Compare the preview with the exact prompt that produced it.
- Copy the prompt and replace only the details that matter for my test.
- Run the new version, then make a small, deliberate change.
Browsing and copying the examples is free. If you use Soutine to generate new images, those generations use Soutine credits. I’m calling that out because a prompt library and an image-generation service are related, but they are not the same thing.
The companion reference package is also available as soutine-banana-prompts on GitHub. It gives the collection a more inspectable home for people who prefer looking at a small source package alongside the live browsing experience.
A small iteration habit
When a prompt is almost right, I try not to rewrite everything at once. I change one variable at a time: the subject, then the lighting, then the composition, then a constraint such as background cleanliness or object count. For example, I might keep a product-shot prompt unchanged while moving from soft window light to a hard studio rim light. That makes the result easier to compare and helps reveal which words actually matter.
If you experiment with Nano Banana, I hope the library helps you get past the blank page and into faster, more controlled iterations. I’m especially interested in seeing which prompt structures make your own results more repeatable.
Top comments (0)