Most disappointing generations aren't the model's fault — they're under-specified prompts. Here are a few habits that reliably raise your hit rate with Nano Banana Pro, the Gemini-based model in Banana Lab.

1. Lead with the subject, then the scene

Models weight the start of your prompt heavily. Name the subject first, then layer on setting, lighting, and mood:

A weathered brass compass resting on a nautical map, soft window light from the left, shallow depth of field, warm tones.

2. Be specific about style

"Make it nice" gives the model nothing to work with. Borrow vocabulary from photography and illustration instead — 35mm, isometric, flat vector, golden hour, studio lighting. Specific words narrow the space the model searches.

3. Use input images when you have a reference

Nano Banana Pro accepts input images as reference or edit material. If you have a composition, palette, or product shot in mind, attach it rather than trying to describe it in words. (The other model, GPT Image 2, is text-only — switch to Nano Banana Pro when you want to pass images in.)

4. Iterate in small steps

Change one thing at a time — lighting, then framing, then palette. It's much easier to see what each edit did than to rewrite the whole prompt and guess.

Found a prompt that works? Share it to the community feed so others can build on it. New to the app? Start with the guide.