To create an AI background for a product photo, start with an accurate photograph, isolate the product, and generate the setting around it. Compare the result with the original before using it. A convincing room or tabletop is useful only if the image still shows the item a customer will receive.
For ecommerce brands and agencies, the bigger task is repeating that process across products and campaigns. This guide covers a practical sequence: approve the product cutout, write a scene brief, test a few variations, review the details, and then reuse the approved setup. The prompts below are starting points to adapt, not examples of tested results.
1. Protect the Product Before Changing the Background
Choose a source photo that shows the correct variant, packaging, and included accessories. It should be sharp enough to inspect the label and edges at the intended display size. Keep an untouched copy as the reference for every review.
Create a cutout or mask that separates the product from its surroundings. Check the rim of a bottle, the spaces inside a handle, thin straps, and small pieces that automatic removal might miss. BasedLabs' guide to making image backgrounds transparent covers the removal and export stage. Save a transparent PNG when moving the cutout between editing tools.
Use an editing mode that restricts generation to the background, if available. Check which part of the mask your tool treats as editable. If you cannot reliably protect the product, generate an empty scene separately and composite the original cutout above it on a locked layer.
Writing "keep the product unchanged" does not enforce that separation. A generative edit can still alter a cap, straighten a logo, invent a seam, or shift the package color. Reference-image support helps communicate appearance; it is not a guarantee that the original pixels remain intact.
Glass and reflective metal need additional attention. A cutout may retain reflections or background colors from the original shoot. Test these products early, and use manual retouching or another photograph when the new setting cannot be reconciled with their appearance.
2. Write a Scene Brief You Can Reuse
Decide where the image will appear before describing the scene. A product detail page, square social post, and tall ad may need different framing. Write down the intended crop and leave space for any text you will add separately.
A useful brief specifies five things:
- Surface: where the product rests, such as pale stone or matte painted wood.
- Lighting: the direction, softness, and warmth that match the source photograph.
- Camera: a viewpoint compatible with the product's existing angle.
- Background: a small number of elements and their distance from the subject.
- Composition: product position, surrounding space, and areas that must stay clear.
Match the scene to the photograph. A bottle photographed straight on will look awkward on a tabletop viewed from directly above. Similarly, a hard highlight on the product is difficult to reconcile with a scene described as evenly overcast.
Use a prompt such as one of these, with the product protected by your chosen editing method:
Skincare bottle: "Create a pale limestone vanity behind the bottle. Soft window light from the left, matching the source photo. Eye-level camera, distant neutral wall, subtle contact shadow. Keep the area around the label clear. No additional bottles or lettering."
Ceramic mug: "Place the background on a light oak kitchen table at the existing camera angle. Diffused morning light from the right. An out-of-focus kitchen beyond the table, no nearby props. Leave clear space above and to the left of the mug."
Headphones: "Create a matte charcoal desk with a warm gray wall behind it. Match the source image's front three-quarter view and upper-left light. Minimal scene, subtle grounding shadow, no cables, screens, or extra accessories."
Keep the approved brief as a template. Change a defined field, such as the surface, instead of rewriting the whole scene each time. Avoid adding decorative ingredients or accessories that could imply they are included with the product.
3. Test a Small Set and Review It Against the Original
Start with one representative product and a few background variations within your budget. Change one major choice at a time. Comparing three surfaces under the same lighting is easier than comparing unrelated scenes with different crops, props, and camera angles.
If you want to connect prompt preparation and image generation, Wireflow's product-photo workflow describes using an existing product photo as a reference and connecting image steps on a visual canvas. Its published example produces studio and lifestyle treatments. Evaluate any output against your original photograph; using a reference image does not remove the need to check product accuracy.
Review at full size first, then at the size shoppers will see. Use the original photo beside the candidate and check:
- Identity: the right SKU, color, proportions, and number of components.
- Text: unchanged spelling, logo, label layout, and visible measurements.
- Edges: no clipped handles, missing corners, halos, or leftover background.
- Placement: believable contact with the surface, without floating or sinking.
- Light: compatible highlights, shadows, reflections, and color temperature.
- Composition: a clear product focal point and enough room for the final crop.
Reject a beautiful image if it changes the item. For a background defect, revise the scene or repair that area while protecting the approved product. For a product defect, return to the original cutout or mask instead of repeatedly editing the damaged version.
Record why you accepted the winning sample: for example, "label unchanged, handle complete, right-side light matches, room for square crop." An agency should also obtain the client's approval of the visual direction before applying it across the catalog.
4. Expand Only After the Sample Is Approved
Before processing a whole collection, test the approved treatment on a small group with different shapes and materials. Include a tall package, a wide product, and a reflective or transparent item if those appear in the catalog. A composition that works for one jar may crowd a taller bottle.
Separate the settings that stay fixed from those that vary. Keep the scene, lighting, crop, and prompt version consistent. Change the source image, product identifier, and any necessary placement adjustment. Store the approved cutout with its source so a teammate can trace an output back to the correct item.
For larger runs, a structured-input workflow can reduce repeated setup. Wireflow's guide to batching approved product-photo briefs describes mapping CSV or JSON rows to generation inputs. Prepare those rows from the approved sample, with fields for the SKU, source image, scene version, aspect ratio, and output name.
Keep a simple status for each image: waiting, generated, rejected, or approved. Save the rejection reason alongside the filename. This lets you rerun failed items without losing track of images that already passed review.
Batch processing repeats instructions; it does not establish that every image is correct. Review each final asset before delivery, and pause expansion if the same defect keeps appearing. Fix the source, mask, or brief before producing more versions of that problem.
5. Prepare Each Image for Its Actual Placement
Keep the editable master and export a separate file for each destination. Check that destination's current image specifications rather than assuming one resolution or background suits every listing. A lifestyle image and a marketplace's primary catalog image may have different requirements.
Inspect the exported file, not just the editor preview. Cropping can remove a handle, compression can blur small print, and a different layout can put text over the product. Upscaling may increase pixel dimensions, but it does not verify or recover the true wording of an inaccurate generated label.
Use descriptive filenames such as mug-sand-oak-scene-v2-square.jpg, and write alt text describing what is visibly present. Deliver approved exports with their product IDs and scene versions. Keep rejected images out of the delivery folder so they cannot be mistaken for finished work.
Frequently Asked Questions
Can I change the background without changing the product?
The most controlled approach is to retain the photographed product on a protected layer and generate the background separately. Masked editing may also help, depending on the tool. In either case, compare the finished image with the source, especially around edges and printed details.
What is the best starting photo?
Use a sharp, correctly exposed image of the exact product variant, with its edges and label visible. Choose an angle and lighting direction that suit the intended scene. A new source photograph is often more useful than trying to rescue an unsuitable one through repeated generation.
How do I stop the product from looking pasted in?
Match the background's viewpoint and light to the original photograph. Check the contact shadow where the product meets the surface, along with scale and reflected colors. If matching the scene requires changing the item's actual appearance, revise the background or choose a more suitable source.
Can I use the same prompt for every product?
Reuse the scene instructions, but test different product shapes and materials before expanding. Tall bottles, wide boxes, and clear glass may need different placement or masking. Save those adjustments as named variations of the approved template instead of making unrecorded changes.
How many variations should I generate before batching?
Begin with a small set that answers a specific question, such as which surface fits the brand. Three candidates can be a manageable first comparison, but there is no required number. Expand after a sample passes product checks and creative review, with a defined budget for the next group.
Custom AI backgrounds become useful production assets when the team can explain what was preserved, what changed, and who approved the result. Keep the real product as the reference, save the scene instructions, and scale the process only after the images pass review.