How to Photograph Clear Products for Background Removal

Clear glass, acrylic and translucent packaging often disappear at the exact moment a background-removal tool is meant to isolate them. The problem usually starts before the software: a clear product photographed against white may not contain enough contrast for its rim, handle, base or liquid line to be identified reliably.

The practical fix is to photograph the product for edge information first. Use a large diffused backlight, place black cards just outside the frame to draw controlled dark edges, and light an opaque label separately. Then create the cutout, inspect its transparency over both light and dark backgrounds, and only use the approved asset in AI scenes.

Why clear products are difficult to cut out

An opaque object has a visible boundary because its pixels differ from the background. A clear bottle is different. Its appearance is partly inherited from whatever sits behind and around it. Reflections show the studio, while refraction bends the background through the material.

That makes transparent-object segmentation a genuinely difficult vision problem, not simply a weak-tool problem. Research behind the Trans10K dataset describes transparent objects as challenging because their appearance is inherited from the background and can resemble their surroundings.

Automatic background removers can still be useful. Adobe says Firefly can trace fine edges and semi-transparent materials, but it also recommends starting with clear contrast between subject and background. The source photograph therefore determines how much reliable edge evidence the tool receives.

Build a bright-field setup for dark, readable edges

For a white or pale listing image, start with bright-field lighting. Place a large diffusion panel, frosted acrylic sheet or evenly lit white background behind the product. Leave enough distance between the product and the panel to position flags without showing them in the frame.

Put the camera on a tripod and keep it level with the product. Lock focus, exposure and white balance after the setup is stable. Clean the item carefully because backlight can make dust, chips and fingerprints more visible.

Now place a black card on each side of the product, just outside the camera frame. Move the cards towards the product until their reflections or refracted shapes create a narrow dark contour along both sides. The cards are not a background. They are edge controls.

Check more than the outer silhouette. Look at the lip, shoulder, base, handle openings, pump tube, liquid line and clear cap. If a boundary vanishes, adjust the card position or add a smaller flag. Do not solve a missing edge by darkening the whole image.

Edmund Optics' backlighting guidance explains the trade-off clearly: backlight creates strong edge contrast and can reveal liquid levels in transparent vials, but it can lose surface detail on the front. That is why the label should be treated as a second lighting problem.

Shoot the label as a separate exposure

A paper label, printed logo or opaque closure usually needs front or side light. Adding enough front light to read the label in the edge frame can wash out the glass contour and fill the product with pale studio reflections.

Keep the camera and product locked. First capture the edge frame with the backlight and black cards. Then add a soft, restrained light for the label and take a second frame. If necessary, make a third frame for a reflective cap or metallic pump.

Composite only the opaque label area into the edge frame. Because the camera and product have not moved, alignment should be straightforward. This is safer than asking an AI editor to reconstruct small text after it has become faint or distorted.

For glossy labels, the separate-frame method pairs well with the cross-polarisation workflow described in Lustra Studio's guide to shooting label detail and surface shine separately: https://lustrastudio.co/blog/cross-polarisation-glossy-product-photography

Remove the background, then test the alpha channel

Run the approved composite through your background-removal tool and export a PNG with transparency. A product on a white JPEG is not the same as a reusable transparent cutout. The PNG needs a usable alpha channel around the product and, where appropriate, through the clear material.

Do not judge the result only on a checkerboard. Place the cutout over white, mid-grey and near-black layers. White reveals dark halos, black reveals missing pale edges, and grey exposes uneven transparency.

Inspect the image at full size and again at normal listing size. Check these areas:

• the complete outer rim, including narrow necks and handles

• the base thickness and contact point

• liquid level and internal tubes

• embossed logos, mould lines and clear caps

• colour fringing or a strip of the original background

• holes that were incorrectly filled or transparent parts made opaque

If the tool has deleted a genuine edge, return to the source and improve the contrast rather than repeatedly regenerating the missing geometry. Lustra Studio's earlier reusable-cutout workflow explains how an approved PNG can then be reused without redrawing the SKU: https://lustrastudio.co/blog/reusable-product-cutout-ai-backgrounds-photoroom

Use the cutout carefully in AI scenes

An opaque cutout can move between backgrounds with relatively predictable edges. A transparent product carries more of the original scene inside it. The background seen through a glass bottle, acrylic box or clear gemstone may still contain brightness and colour from the studio setup.

When the replacement scene differs greatly, a simple cutout may look pasted on. The internal refraction will not automatically match the new surroundings, and generated reflections can change the perceived shape or material. Use subtle backgrounds first, keep the product itself outside the generative selection, and compare the output with the source.

For difficult products, create a light-background master, a darker edge-defined version and the untouched original. The guide on choosing a transparent or white AI reference explains when each source is more dependable: https://lustrastudio.co/blog/transparent-png-vs-white-background-ai-product-reference

Know when a real photograph is still required

Do not use a generated background to prove optical performance, liquid colour, material clarity, wall thickness or the contents of transparent packaging. Photograph those claims directly. Clear eyewear, lenses, measuring vessels and products with safety-critical liquid levels need especially careful treatment.

If the item contains etched markings, tiny bubbles, inclusions or deliberate optical effects, automated removal may classify them as background noise. Preserve a high-resolution original and compare every edited file with it.

The aim is not a perfectly invisible background at any cost. It is a reusable product asset whose silhouette, transparency, label and material still describe the real item. Give the software edges it can see, test the alpha channel on contrasting colours, and reject any cutout that becomes cleaner by deleting product information.

Sources

Edmund Optics, “Backlight Illumination for Machine Vision,” reviewed 20 October 2026: https://www.edmundoptics.com/knowledge-center/application-notes/illumination/backlight-illumination-for-machine-vision/

Edmund Optics, “Common Illumination Types,” reviewed 20 October 2026: https://www.edmundoptics.com/knowledge-center/application-notes/illumination/choose-the-correct-illumination/

Adobe Firefly, “Free Transparent Background Remover,” reviewed 20 October 2026: https://www.adobe.com/products/firefly/features/remove-background/transparent.html

Xie et al., “Segmenting Transparent Objects in the Wild,” submitted 31 March 2020: https://arxiv.org/abs/2003.13948