Image-to-Image Editing

Image-to-image lets you give the model a starting picture and a description of what to change. It’s the workhorse for retouching, restyling, swapping backgrounds, dressing characters, and more.

Setup

1. Add an Image node and upload your source image
2. Add a Prompt node describing the change
3. Add an Output node and connect both:
   • Prompt → output’s prompt handle
   • Image → output’s images handle
4. In the Output node, pick a model that supports image-to-image (Nano Banana 2 is the default)

Writing the prompt

For edits, describe the change, not the whole image:

✅ “Replace the sky with a stormy sunset, keep the foreground unchanged.”

❌ “A coastal landscape at sunset with a stormy sky and a small lighthouse on the right.”

Models trained for editing pay attention to what’s in the source image; they don’t need you to describe it again.

Multi-image fusion

Some models accept multiple input images. Add a second Image node and connect both into the same Output node’s images handle. Useful for:

• Style transfer (image A’s content + image B’s style)
• Compositing (placing one subject into another scene)
• Outfit / character swaps

Iterating non-destructively

When you click Edit on a generated image and save, Banana Flow creates a new Image node — it doesn’t overwrite the previous output. This means you can branch and explore variants without losing prior steps.

Tips

• Lower-cost models like Nano Banana 2 are great for rapid iteration; switch to Pro/Ultra only when the composition is locked in.
• Keep source images at a reasonable resolution. Very large inputs slow down generation without much quality benefit.
• For exact-region edits, use the in-canvas editor’s mask/draw tools first, then send to image-to-image.