How to Edit an Image Using an AI Reference

Updated November 2, 2026

A reference image can show the style, lighting, or detail you want without replacing the subject in your original image. The key is to tell the editor which image to preserve and exactly what to borrow from the reference.

Prepare your source and reference

Choose a clear source image with the subject visible and enough resolution for your intended output. Pick a reference that demonstrates one or two specific qualities, such as soft studio lighting or a muted color palette. You should have permission to use both images.

Decide what must stay unchanged before generating: for example, the product shape, label, camera angle, and background layout. Choose an aspect ratio for the destination, such as 1:1 for a square post, 4:5 for a feed image, or 9:16 for a vertical story. Changing the ratio can require cropping or extending the composition.

Edit the image in four steps

1. Open the editor and upload your source. Go to Oranzy and select an image-editing workflow. If you are new, sign up for 100 bonus tokens. Upload the image you want to change and choose an editing model that supports image inputs.
2. Add the reference. Upload the reference as a second image if the selected workflow supports multiple image inputs. Identify the first image as the source and the second as the reference. If the workflow accepts only one image, use the source as the input and describe the reference's visible qualities in your prompt instead.
3. Write the edit request. Name the change, what to preserve, and the intended aspect ratio. For example: "Edit the first image of the ceramic bottle. Keep its shape, label, camera angle, and position unchanged. Use the second image only as a reference for soft side lighting and a warm beige palette. Keep the background simple. Output in 4:5."
4. Generate, compare, and refine. Review the result against both inputs. If needed, generate another version or revise one instruction at a time. For example, add "keep the original label text unchanged" if the label has drifted, then compare the new result with the previous one.

Compare results and refine the prompt

Check each version at the size it will be used. Compare the subject's shape and position with the source, then check the requested style or detail against the reference. Inspect small text, edges, reflections, and background objects at full size; generated text and fine product details may be inaccurate.

Use a specific correction rather than rewriting the whole prompt. If the color is right but the lighting is too harsh, try: "Keep the current warm palette; soften the shadows and reduce contrast." If the editor allows you to use a generated result as a new input, refine the strongest version rather than starting over. Keep the original source so you can return to it if details drift.

Avoid common mistakes and wasted generations

• Vague instructions: Replace "make it look like the reference" with the exact lighting, color, texture, or detail you want to transfer.
• Conflicting directions: Do not ask for an unchanged background and a completely new setting in the same prompt. Prioritize the elements that matter most.
• Changing too much at once: Test a short prompt first, then adjust one property per iteration so you can see what caused the difference.
• Using a poor reference: A blurry or cluttered image makes the desired feature harder to identify. Crop the reference to the relevant area when appropriate.
• Trusting fine details without checking: Verify labels, logos, and product features manually. For exact text or branding, finish with a conventional image editor if the generated version is inaccurate.
• Ignoring output format: Set the intended aspect ratio early and inspect the final crop before publishing.