Use this guide when you have generated an image that is close, but you want to change a specific part of it.
The edit box in Preview is not a chat. It is a prompt box for an AI image editing model. That means it works best when you give it a clear visual instruction: what should change, what should stay the same, and which reference images it should use.
When to use Edit
Use Edit when:
the overall image is already good
you want to change one area of the image
you want to remove, replace, or adjust a visible detail
you can describe the final result clearly
you have a useful reference image for the thing you want to match
Edit is good for changes like:
making lighting warmer or cooler
removing a small object
changing a prop
adjusting a background detail
fixing one visible part of a product
applying a reference detail to an existing image
If the whole image is wrong, it is usually better to regenerate or remix instead of trying to fix everything with one edit.
How to use the masking tool
Masking tells Riverflow which part of the image you want to edit.
Open the image you want to change.
Click Edit.
Select the pen tool.
Draw over the area you want Riverflow to change.
Attach any relevant reference image.
Write a clear edit prompt.
Click Generate.
Mask the full area that needs to change. If you are replacing an object, cover the object and a small margin around it so the lighting, edges, and shadows can blend naturally.
Masking helps Riverflow focus the edit, but it is not a pixel-perfect selection tool. If the edit needs exact object placement, exact typography, or precise design layout, you may need to use a design tool or ask our team for help.
How to write a good edit prompt
A strong edit prompt has three parts:
What should change.
What should stay the same.
Which reference image to use, if you attached one.
Use this structure:
Change only [specific area or object] to [desired final result].
Preserve [important parts of the image that should not change].
Use the attached reference image for [specific detail] only.
Do not change [anything important to protect].
Good edit prompts are direct and visual.
Instead of:
That is still not what my product looks like. Copy the look and detail from the product photo exactly.
Use:
Change only the product to match the attached product reference.
Preserve the scene, camera angle, lighting, shadows, background, and product placement.
Keep the product shape, proportions, label placement, colours, material finish, logo, and visible text faithful to the reference.
Do not redesign the product or change the scene.
Good prompt examples
Make a small visual change
Make the lighting warmer and softer.
Preserve the product, camera angle, composition, background, and all visible text.
Remove an object
Remove the cup from the selected area.
Fill the background naturally so it matches the surrounding table surface, lighting, and shadows.
Do not change the product or the rest of the scene.
Replace a prop
In the selected area, replace the orange slices with sliced pear.
Match the existing camera angle, lighting, shadows, and image style.
Do not change the product, label, background, or composition.
Match a product reference
Change only the product to match the attached product reference. Preserve the current scene, camera angle, lighting, shadows, and product position. Keep the product shape, proportions, label placement, colours, material finish, logo, and visible text faithful to the reference. Do not copy the reference image background.
Correct product text or label details
Change only the product text and label details that are incorrect.
Preserve the product geometry, proportions, cap shape, camera angle, lighting, shadows, label boundaries, material finish, and background.
Replace the visible product text with exactly: "[insert exact text]".
Do not redesign the product or change the scene.
For more detailed product text guidance, use How to Improve Product Text Accuracy.
How to use reference images well
The editor does not remember what you mean by "my product", "the original", or "the reference" unless you attach the relevant image.
If you want Riverflow to match a product, logo, label, texture, person, prop, or style, attach a reference image and explain what it should be used for.
For example:
Use the attached product reference for product shape, label placement, colour, logo, and material finish.
Do not copy the reference image background, camera angle, or lighting.
Use reference images for:
product shape and proportions
label or artwork details
logo placement
material finish
prop appearance
human appearance or styling
lighting or mood
texture or surface detail
Try to use the clearest reference possible. A good reference image is:
high resolution
sharp and well lit
focused on the detail you want to match
not crowded with unrelated objects
not a full moodboard unless you explain exactly what to borrow from it
If you attach more than one reference image, give each one a job.
Use image 1 as the base image.
Use image 2 as the product reference.
Use image 3 for lighting style only.
Do not copy the background or props from image 3.
What to avoid
Avoid chat-style feedback.
No, that is still wrong. Try again and make it more like the photo.
The editor does not reason through the conversation like a support agent. Rewrite the instruction as a fresh image-edit prompt.
Avoid vague prompts.
Make it better.
Make it premium.
Fix the product.
Improve the image.
Use specific visual language instead.
Make the background a warm cream colour.
Remove the shadow on the left side of the label.
Make the product label match the attached reference.
Keep the bottle shape and cap unchanged.
Avoid stacking too many edits into one prompt.
If you ask Riverflow to change the product, background, props, lighting, crop, text, and hand position all at once, the edit is more likely to drift.
Work in smaller steps:
Fix the most important issue first.
Choose the best result.
Make the next edit from that result.
Avoid relying on text prompts for highly precise layout changes.
Some edits are surprisingly challenging for AI image editors, especially:
moving an object to an exact position in 3D space
extending a background while keeping every object unchanged
changing text while preserving all packaging details
editing scenes with many products or overlapping objects
changing product size without affecting shadows, reflections, or perspective
For those cases, use a stronger reference image, mask the exact area, simplify the request, or contact support before spending too many credits.
Which tool should I use?
Use Edit when the image is mostly right and you want to change one specific area.
Use Fix Product Details when the product artwork, label, logo, or visible text needs to be more accurate.
Use Swap Product when the scene is good but the product itself is wrong.
Use Regenerate or Remix when the composition, camera angle, scene, or overall concept is not close enough.
Use Generate Angles when you want a new view of the product rather than a small edit to the same image.
Which edit model should I use?
For most edits, start with RF2.5 Pro High.
This should be your default edit model. It is usually the best balance of quality and credit usage for:
product text and label details
product accuracy
reference-based edits
masked edits
most general image edits
Use RF2.5 Pro x High more sparingly. It is the most expensive edit model, so it is best saved for very detailed work where accuracy matters most, such as difficult product details, complex references, or edits where the first attempt with RF2.5 Pro High is not good enough.
If your edit involves visible text, labels, logos, or product detail, start with RF2.5 Pro High. Move to RF2.5 Pro x High only when you need the extra quality.
GPT Image 2 is also a strong option for text-focused edits. If you are trying to improve product wording, label clarity, or typography, it is one of the top models to test.
The top three edit models are:
RF2.5 Pro High - default choice
RF2.5 Pro x High - highest quality, highest credit cost
GPT Image 2 - strong for text-focused edits
RF2.0 and NB2 can be useful for softer stylistic edits, such as lighting, mood, colour tone, or atmosphere. However, if you are unsure which model to use, choose RF2.5 Pro High.
If you are on the Free plan, you may need to subscribe to access the more advanced edit models.
Quick checklist
Before you generate an edit, check:
Have I said exactly what should change?
Have I said what should stay the same?
Did I attach the right reference image?
Did I explain what the reference image should be used for?
If using a mask, did I cover the full area that needs to change?
Am I asking for one main edit, rather than five edits at once?
Is this a case where Fix Product Details, Swap Product, or Regenerate would work better?