Nano Banana 4
← Back to guides
Updated 8 min read

Photo Editing with Nano Banana 4: Retouch, Restore and Restyle by Talking

A hands-on tutorial for conversational photo editing: object removal, background swaps, outfit changes, old photo restoration and multi-step edit chains.

Nano Banana 4 replaces an afternoon in Photoshop with a few sentences. This tutorial walks through the five most useful editing workflows, with copy-paste instructions for each.

How conversational editing works

Upload a photo, then describe the change. The model re-renders the image with your edit applied while preserving everything else. Two rules govern good results:

  1. Name the target precisely. "Remove the red car on the left", not "clean up the photo".
  2. State what must not change. End edits with "keep everything else exactly the same" β€” this is the single most useful phrase in photo editing.

Workflow 1: Remove objects and people

Remove the two tourists in the background and the power lines
in the sky. Keep the main subject, lighting and colors
exactly the same. Fill the removed areas naturally.

Works on photobombers, trash bins, wires, watermarks and reflections. For crowded backgrounds, remove in two or three passes rather than all at once β€” each pass gives the model a cleaner canvas.

Workflow 2: Swap the background

Replace the background with a quiet beach at dusk, soft warm light.
Keep the person completely unchanged β€” same pose, same expression,
same clothing. Match the lighting on the person to the new scene.

The last sentence is the professional touch: without "match the lighting", subjects look pasted on. With it, the model relights hair edges and skin tones to blend with the new environment.

Workflow 3: Change outfits

Change the outfit to a tailored charcoal suit with a white shirt.
Keep the face, hairstyle, body pose and background unchanged.
Make the fabric drape naturally with realistic folds.

Describe fit and fabric ("tailored", "linen", "oversized knit") β€” generic clothing names produce generic results. This workflow powers virtual try-ons, professional headshots from casual photos, and costume tests.

Workflow 4: Restore old photos

Restore this old damaged photograph: repair scratches, tears
and faded areas, sharpen facial details, then colorize with
natural colors accurate for the 1960s. Preserve the original
faces and expressions exactly.

Two tips from experience: state the decade so clothing and color tones stay period-correct, and for badly damaged photos run restoration first, colorization second as separate steps β€” combining them on a heavily damaged scan can hallucinate details.

Workflow 5: Chain edits like a pro

Real editing is rarely one step. Chain small, single-purpose edits:

  1. "Remove the crowd in the background"
  2. "Make the lighting golden hour"
  3. "Slightly sharpen the subject's eyes"
  4. "Crop to a vertical 4:5 portrait composition"

Each message edits the latest result. If a step goes wrong, say "undo that last change and instead..." β€” the conversation is your history panel.

When the edit fails

  • The subject's face changed β€” add "keep the face pixel-identical" and retry from the previous step.
  • The edit was ignored β€” your target description was ambiguous; name its position and color explicitly.
  • The result looks pasted β€” ask to "match lighting, shadows and grain across the whole image".

For maintaining one person's identity across many edited images, continue with the character consistency guide.