Back/Content/Midjourney
BeginnerContent

How to Fix and Refine AI-Generated Images

Repair one weak detail in a strong AI image by selecting the problem area and describing the replacement with enough spatial constraints to preserve the composition around it.

How to Fix and Refine AI-Generated Images

Jamey treats Nano Banana like conversational Photoshop, replacing an incorrect computer while preserving its angle, size, composition, and the rest of the Midjourney image.

Before you start

What you need

  • A generated image worth keeping
  • Nano Banana, Flora, or another targeted image editor
  • A precise description of the replacement object

What you’ll make

A higher resolution image with the faulty object corrected and the surrounding scene unchanged.

Tools used

Step by step

The workflow

Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.

3 steps

Step01

Upload Image to an Editing Tool

Choose a generated image whose composition is already strong. Upload the highest quality version to Nano Banana through Flora or another editor and define the one defect you want to change.

Step02

Mask the Area to Change

Mask only the incorrect object or feature. Leave enough edge context for the model to match lighting and perspective without exposing the entire image to regeneration.

Step03

Prompt the Change with Specific Instructions

Name the replacement precisely, describe its visible surfaces and orientation, and lock the position, size, subject, background, and composition. Inspect the edit at full resolution before exporting.

Example prompt
Replace the computer she is typing on with a 2026 midnight black MacBook Pro. Keep the computer in exactly the same position and size. Only its left side and keyboard are visible. Preserve the woman, pose, hands, lighting, background, crop, and every other part of the image. Upscale while retaining natural fabric and skin texture.

Add a constraint like 'don't change anything else' to your prompt to ensure the rest of the image stays consistent during the edit.

What good looks like

  • The replacement object is recognizable and current.
  • Its size, angle, and visible surfaces fit the original pose.
  • Lighting, texture, and composition outside the edit remain stable.
  • The exported image has enough resolution for its intended use.

Build your next product with ChatPRD

Turn an idea into a PRD, user stories, and a plan.

Try ChatPRD free

After the steps

Runbook notes

How to recover when the loop fails and where human judgment helps.

Recover

If it goes sideways

The editor changes the subject or background
Tighten the mask and state that every unselected region, face, pose, lighting choice, and background must remain unchanged.
The replacement object has the wrong perspective
Describe which sides are visible, the object orientation, and its exact position and size in the frame.
Upscaling removes texture or makes the image plastic
Request texture preservation, compare at full size, and separate the object edit from the final upscale if needed.

Start shipping
better products.

Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.

Free to start
No credit card
SOC 2 certified
Enterprise ready