Nano Banana is Google's image model family, available on Higgsfield: four models covering generation, editing, and batch production. The most used are Nano Banana Pro for precision generation with prompt reasoning, Nano Banana 2 for fast production at scale, and Nano Banana for editing existing images by instruction. The table below shows every model and what it's best for.
What is the Nano Banana family?
Model | Best for | Key detail |
|---|---|---|
Nano Banana | Edit an existing image with precise instructions | Up to 8 reference images |
Nano Banana Pro | Complex scene generation with reasoning | Analyzes the prompt before rendering, up to native 4K |
Nano Banana 2 | Fast bulk generation at Pro-level quality | Up to 5 consistent characters + 14 stable objects |
Nano Banana 2 Lite | Fastest tier for high-volume drafts and batch production | ~4 seconds per image, 1K resolution |
Which Nano Banana model should I use?
Nano Banana Pro is built around a reasoning core: it analyzes the prompt before rendering, which gives it the highest precision with quantities, text, faces, and perspective. Nano Banana 2 delivers Pro-level quality at higher speed and holds up to 5 consistent characters in one scene. Nano Banana 2 Lite is the fastest tier, built for volume: around 4 seconds per image at 1K. Nano Banana is the editing model: it changes existing images by instruction rather than generating from scratch.
How do I generate with Nano Banana Pro?
Open higgsfield.ai → Image → Nano Banana Pro.
Choose your input mode: Text-to-Image, Multi-reference, or Image-to-Image.
Write a clear, specific prompt. Describe the subject, setting, lighting, and any specific quantities or text. Be explicit.
Add references if needed (up to 14), and describe the role of each reference in the prompt.
Set the resolution. Start at 1K when iterating, scale to 2K or up to native 4K for final assets.
Generate and iterate. If something is off, change the prompt or the reference, not both at once.
Nano Banana Pro analyzes the prompt before rendering: it evaluates relationships between objects, checks for logical consistency, and interprets quantities, spatial layout, and lighting. This is what keeps the number of objects correct, text readable, faces consistent, and perspective accurate.
How do I generate with Nano Banana 2 and Nano Banana 2 Lite?
Open higgsfield.ai → Image and select Nano Banana 2 or Nano Banana 2 Lite in the model picker.
Choose your input mode: Text-to-Image, Multi-reference, or Image-to-Image.
Write a clear, specific prompt. The same rules as for Nano Banana Pro apply.
Set the output. On Nano Banana 2, pick the resolution (up to native 4K) and the aspect ratio. Nano Banana 2 Lite generates at 1K: pick the aspect ratio and a Thinking level (High or Minimal) in place of a resolution picker.
Generate and iterate. Draft at scale on Lite, then rerun the final version on Nano Banana 2 or Pro when resolution or maximum precision matters.
How do I edit an image with Nano Banana?
Open higgsfield.ai → Image → Nano Banana.
Upload or create a base image.
Write a short instruction: "Change the background to a Parisian café at night," "Replace the T-shirt with a black leather jacket," "Insert text: FUTURE IS NOW."
Add references if needed: up to 8 images in one composition, each with its own role described in the prompt.
Generate.
Nano Banana understands product and cultural context, keeps characters consistent between frames, and can add or edit text on images.
How do I keep a character consistent across generations?
Save the character as an Element: click @ Elements under the prompt field (or type @ in the prompt) and select it, with no re-uploading per image. If the character is a real person, train a Soul ID first and create the Element from its portraits. See How do I create and use a Soul ID character?
How do I get more accurate outputs?
Output ignores a specific quantity or layout. Be more explicit: "exactly 3 bottles on the left side of the frame" rather than "a few bottles."
Text in the image is blurry or misspelled. Put the exact text in quotes in your prompt, and describe the font style, size, and placement explicitly.
Output looks blurry at 2K or 4K. Retry the generation. If it persists, generate at a lower resolution and upscale the result.
Reference image is rejected. Check that the file is JPG, PNG, or WebP and within the size limit, and avoid images with heavy compression or watermarks.