Remove Backgrounds From Product Images With AI: Glass, Hair, Shadows

Shoot against a plain, contrasting backdrop, run the photo through a background remover, then repair the three places it fails: hair or fur (soften and defringe the mask), glass (shoot on white and blend, or rebuild the transparency on its own layer) and shadows (keep a contact shadow on a separate layer). Test the cutout on a dark colour before you export.

The one-click tools are good at solid objects with clean outlines: a mug, a box, a shoe. They are not good at what makes products look real. Fur and fine hair carry thousands of semi-transparent pixels; glass lets the old background show through the product; shadows sit between the product and the floor and belong to neither. These are the edges that make a listing look cut out with scissors, and they are the cases the steps below concentrate on.

Follow me on Instagram@sagnikteaches

A repeatable routine has five parts: a short stress test that shows what a tool can and cannot do, a fix for each hard edge, marketplace-ready exports, a batch run, and a check at the end. The tools named here are options rather than recommendations, and their prices and limits change, so read the current page before you commit.

Connect on LinkedInSagnik Bhattacharya

Why edges fail, in plain terms

A cutout is a mask: a second image that says, for every pixel, how much of the photo to keep. White means keep, black means discard, grey means keep partly. A hard-edged object has a mask that is nearly all black and white. Hair, fur, mesh, a cotton jumper's fuzz and the rim of a wine glass need grey, and many tools make that grey badly, so you get a hard, jagged edge or a soft halo of the old background.

Subscribe on YouTube@sagnikteaches

The second cause is colour spill. The edge pixels were photographed with a bit of the old backdrop mixed in, so a dark rim round a pale product, or a white rim round a dark one, appears when you place it on a new colour. The tool's job is to remove that contamination; the cheap ones do not.

The third cause is the photo. If the product and the backdrop are nearly the same colour, if the edge is blurred by movement, or if the background is cluttered, no algorithm can tell where the product stops. Fix those at the shoot, and your edge problems shrink by more than any tool change will achieve.

Shoot so the tool has an easy job

• Contrast. A pale product on a mid-grey or coloured backdrop, a dark product on white. If the product is white and the marketplace wants white, shoot on a mid-grey, cut out, then place on white.
• Distance. Put the product 30 cm or more from the backdrop so it does not pick up colour or cast a shadow onto it that merges with the product.
• Sharpness. Use a tripod or a table-top rig. A blurred edge is a blurred mask.
• Size. Shoot larger than you need. A 4,000-pixel photo cut out and shrunk to 2,000 hides edge flaws that a 2,000-pixel original would show.
• Light the backdrop and the product separately if you can. A bright, even backdrop behind a product makes white-backdrop shots easier; it also makes the glass problem below simpler.

If you are comparing shooting set-ups and AI-generated backdrops in general, AI tools for ecommerce photography: backgrounds, shadows and consistent sets covers the wider picture; this tutorial stays on the cutout itself.

Build a stress set before you pick a tool

Do not judge a remover on a mug. Pick ten of your own photos that are hard, and run each candidate tool on all ten. The set should include a fur or fibre item, a glass or clear plastic item, a reflective metal item, a product with a hole (a handle, a ring), a pale product on a pale backdrop, a dark product on a dark backdrop, a product with fine detail such as a chain or mesh, one with a shadow you want to keep, and two ordinary items for comparison.

Score each result out of 5 on four things: outline accuracy, edge softness, colour fringe, and what happened to transparent parts. Save the results to a folder with the tool's name. After an hour you will know which tool fits your catalogue, and you will know it from evidence rather than from the sample images on the tool's home page.

This table is a starting list, not a ranking. The examples in this tutorial work with any of them, since the repair steps are about masks and layers rather than a brand. One detail from Canva's help page is worth knowing if a platform asks you to declare AI edits: it treats its remover as an editing tool and says it does not add Canva's AI-generated content metadata to the design.

Hair and fur: soften the mask, then remove the fringe

An illustrative case: a pet shop photographs a plush fox toy with long, loose fur on a pale grey backdrop. The one-click result has a hard edge that looks cut, and tiny tufts have vanished. The fix has three parts.

1. Use the hair-aware option if the tool has one. Many tools have a mode, slider or refine brush for hair and fur. If yours does not, move the cutout into an editor that does.
2. Soften, do not blur. Add a feather of about 0.5 to 1.5 pixels on the mask, enough to remove stair-stepping but not enough to smear. Zoom in to 200%.
3. Remove the fringe. If a grey halo shows when the toy is placed on white, shrink the mask by 1 or 2 pixels, or use a defringe or "decontaminate colours" step that replaces edge colour with colour from deeper inside the product.

The same steps work for human hair on a model, a woolly jumper and a feather boa. Check the result on a white layer, a black layer and a mid colour. A cutout that looks good on only one of them has a fringe problem you will see the day a listing needs a different backdrop.

A prompt can help you plan the fix, though it cannot see the image. Describe what you see and ask for the order of operations in your tool.

I am cutting out a plush toy with long fur from a grey backdrop in [tool]. The one-click cutout has a hard edge and a grey halo on white. List the steps in order to fix it in [tool], using only features the tool really has. If you are not sure a feature exists in the current version, say so and tell me what menu or help page to check. Do not invent menu names.

That last instruction is the one that matters: assistants invent menu items with great confidence. Treat any feature name it gives you as a thing to confirm in the tool's help page.

Glass and clear plastic: the cutout cannot see through

An illustrative case: an optician photographs a pair of glasses in clear-lens frames. The remover cuts a perfect outline, but the lenses now show the old grey backdrop through them, or go fully transparent and look like holes, or fill with solid white. All three are wrong, for the same reason: a cutout assumes each pixel belongs to the product or to the background, and glass belongs to both.

There are three workable fixes, from easiest to most careful.

1. Shoot on white and blend. Photograph the product on a pure white backdrop with the light set so the background reads as white. Instead of cutting out, place the photo on a white page with a multiply blend, so only the darker parts of the product show. This keeps the glass as it was photographed, because the backdrop it saw was white. It works for white or very light final backgrounds only.
2. Cut out the frame and rebuild the lens. Remove the background, then make a separate layer for each lens: a very pale grey at about 10 to 20% opacity with a small highlight painted or copied from the original. It looks less real than the original, but it reads as glass on any backdrop.
3. Two-shot method. Shoot the product on white, then on black. Use the difference between the two to extract a true transparency mask, which is the careful, studio route. It takes more time than a small shop needs for every item; keep it for hero images.

The same idea suits perfume bottles, drinking glasses, clear packaging, jars and acrylic displays. A cosmetic bottle full of liquid is also better shot against a lit white than cut out and floated. For the colour fidelity issues on cosmetics, cosmetic product photography with AI: textures and honest colour goes deeper.

Reflective items: chrome, mirrors and polished metal

A shiny item reflects the room, which includes the photographer. A cutout of a chrome kettle will carry reflections of a window and a wall that look wrong on a clean white background. Removing the background does not fix that; the reflection is part of the product pixels.

Shoot reflective items inside a light tent or surrounded by white card so that what is reflected is white. Then the item looks right on white and the edges are predictable. If you only have a finished photo with a messy reflection, a few options: retake it; clone out the offending reflection in an editor; or accept a simplified look with a soft white gradient painted in. Do not rely on a generative fill to invent a reflection, because it may invent details that the real product does not have. For small polished items such as rings and watches, jewellery photography with AI: clean backgrounds that stay true to life has more on reflections and colour.

Shadows: keep them, but on a layer of their own

A product on pure white with no shadow floats. A product with the original dark, hard shadow under it looks dirty. The usual answer is a soft contact shadow: a faint, blurred shape directly below the product that shows it sitting on a surface.

Many removers have a "keep shadow" or "add shadow" option; check how yours treats the original. A more controllable route is to remove the old shadow with the background, then create a new one in your editor on its own layer: a flattened copy of the product silhouette, filled with black or dark grey, blurred 20 to 40 pixels, squashed vertically to about a third of the height, set to 15 to 30% opacity and placed under the base. Save the shadow layer separately so you can export two versions: one with the shadow for your own site, and one without for a marketplace that wants a clean white background.

Marketplace rules differ on whether a shadow is allowed on a main image, and they change. Read the current image page of each marketplace you list on. Google's product-image guidance, for example, asks for a clean, unobstructed view of the item rather than a white background, rules out borders, watermarks and promotional overlays, and suggests as a best practice that the product fill 75 to 90% of the frame. Many marketplace main-image rules ask for a plain white background and for the product to fill most of the frame; treat any exact percentage you see quoted by third parties as unconfirmed until you read it on the marketplace's own help page. If you sell on several, multichannel listing with AI across Amazon, eBay and Etsy shows how to meet different rule sets from one source photo.

Marketplace-ready exports: a short settings list

• Background value. If the rule says pure white, check it as a number (255, 255, 255), not by eye. Off-white backgrounds that look white on screen can fail automated checks.
• Canvas. A square canvas with the product centred and an even margin satisfies most grids. Decide a margin (say 5% of the width) and use it for every product.
• Size. Export at the largest size the marketplace accepts or recommends, since many offer zoom only above a size threshold. Look up the figure on its help page: Google, for instance, recommends around 1,500 × 1,500 pixels or more, with a 500 × 500 minimum applying to all products from 31 January 2027.
• Format. JPEG for the white-background version, PNG with transparency for your own site if you want to place products on coloured sections.
• File names. Include the product code and an image number, for example SKU1042_main.jpg, so you can find and replace them later.
• No watermarks, text or borders on main images, unless the marketplace says otherwise. If you add AI-generated parts to an image, check whether the platform asks for the image to be marked as AI-edited; Google Merchant Center image requirements for AI-generated product photos sets out the metadata rule for Google.

The pet shop batch: a worked example with numbers

A pet shop has 180 products to photograph for a new site: 60 toys (about 20 with fur), 50 bottles and tubs (glass and clear plastic), 40 solid items such as bowls and collars, and 30 bags of food. It shoots them in four sessions on a mid-grey backdrop with the same light.

1. Stress set (1 hour). Ten photos run through two candidate tools. One tool handles fur better; the other is quicker on solid items. The shop picks the first and plans to fix by hand where it fails.
2. Batch run (about 40 minutes of waiting). All 180 go through the chosen tool in one batch, keeping the original file names.
3. Check pass (2 hours). A staff member opens each result over a black layer and a white layer, one flip each, 15 seconds per image: 180 × 15 seconds is 45 minutes per layer. She marks the fails in a sheet.
4. Repair pass. Of 180, 123 pass first time (68%). 31 have fur or fringe problems, 19 are bottles with lens-like clear areas, and 7 have shadow or reflection issues. Fur fixes take about 6 minutes each (186 minutes), glass bottles about 8 minutes (152 minutes) and the rest about 5 minutes (35 minutes): roughly 6 hours 15 minutes.
5. Export. Two exports per product: white-background JPEG and shadowed PNG.

Total hands-on time: about 9 hours for 180 products, so around 3 minutes a product. A fully manual job at 10 to 15 minutes a product would be 30 to 45 hours. The saving comes from the stress test and from fixing only the 32% that fail. To automate the later steps of resizing, naming and placing on backgrounds, see product photography automation: batch backgrounds, resizing and file names.

A quality check you can run in five minutes per batch

Looking at a cutout on white hides most of its flaws, because white hides pale fringes. Use this check on each batch, or on a sample of one in five if the batch is large.

1. Place each cutout over a saturated dark colour, such as black or deep red. Pale halos show instantly.
2. Place it over a mid-grey. Dark fringes show.
3. Zoom to 200% on the three hardest edges: the top, the thinnest part and anywhere there is a hole.
4. Look inside glass and holes: is the old background still visible, or has it turned white?
5. Compare the colour of the product to the original, since some tools shift tones slightly.
6. Check the file against the marketplace checklist: pure white value, margin, size, name.

Record failures by type in a small sheet: fringe, hair, glass, shadow, colour, other. After a few batches you will see which type dominates, and you can fix the shoot rather than the cutout. If hair is your main fail, change the backdrop; if glass is, change the lighting.

When a quick fix makes things worse

Picture an optician's shop that cuts out forty frame photos in one go and places them on white. At thumbnail size they look fine. When a customer zooms in on a listing, the clear lenses have turned into solid white ovals with a thin dark edge, so the glasses look as if they have paper in them. The shop approved the batch on the thumbnail grid and never opened one at full size.

It shows up in the first week as a handful of questions asking whether the lenses are tinted. The fix is to reshoot the lens-bearing frames on a lit white backdrop and use the multiply method from the glass section above, leaving the plain-lens frames as they are. The lesson is the check routine: always inspect a sample at full size on two backgrounds, and never approve from a grid.

Where AI helps and where a person must look

The removal step is the part AI does well; use it to save hours. A chat assistant is also useful around the edges: it can draft the checklist for your marketplaces from pages you paste, write a naming convention, build a spreadsheet template for the fail log, or write a short brief for a freelancer who will fix the hard 30%. It cannot judge a cutout it cannot see, and it should not be asked to confirm that an image meets a marketplace's rule from memory.

Keep a person on three checks: the edge inspection on dark and light layers, the product's true colour against the original, and the marketplace rule against its current help page. If you go beyond removal and generate a new scene behind the product, treat the result as an AI-generated image: Google's image guidance says images created with generative AI must carry metadata saying so, and asks sellers not to strip the IPTC DigitalSourceType tag the tool writes.

Background removal questions from small sellers

Why do AI cutouts leave a white or dark halo?

The tool keeps a thin ring of semi-transparent edge pixels that still carry the old background's colour. Placed on a new colour, the ring shows. Fix it by shrinking or contracting the mask by one or two pixels, using a 'remove fringe' or 'defringe' step, or by re-cutting against a background close to the final one.

Can AI remove the background from a glass bottle?

It can cut the outline, but glass is see-through, so the cutout either keeps the old background inside the glass or goes opaque. The reliable fix is to shoot on white or black and use a multiply or screen blend, or add the glass back from a separate layer, then check the result on the final colour.

Should I keep the shadow when I remove the background?

Keep a soft contact shadow for your own website, because it stops the product floating. Marketplace main images often need a clean white background, and rules differ on shadows, so keep the shadow on its own layer and export two versions.

Is it fine to use a free background remover for a whole catalogue?

Free tiers often limit resolution, daily uses or commercial use, so read the current terms for each tool before batching a catalogue. Test ten difficult images first. A free tool that ruins 20% of edges costs more in fixing time than a paid plan costs in a month.

Do I still need good lighting if AI removes the background?

Yes. Good light, a plain backdrop and a sharp subject give the tool clean edges. Most failures are not the AI's fault but the photo's: a busy background, motion blur, or a product the same colour as what is behind it.

Further reads

• How to Create Product Photos With AI on a Small Budget — Build a whole product photo set on a small budget.
• AI Product Photos vs a Photographer: Cost and Quality Compared — Weigh AI photos against hiring a photographer.
• Dropshipping Product Photos: Replace Supplier Images Safely — Replace supplier images with your own cut-outs safely.
• Free AI Product Photoshoot: What Free Plans Really Allow — See what free AI photo plans allow before you batch.
• How to Use AI to Stage Furniture for Product Page Room Sets — Build room sets around your real product photos, check scale, colour and construction against the piece, and label the result honestly.
• Shopify SEO With AI: Fix Titles and Meta Descriptions in Bulk — A CSV round trip for Shopify SEO text: export, rewrite with AI in batches, check length and duplicates, import, and judge it in Search Console.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources (read 5 November 2026): Photoroom's pricing page (Background Remover in Pro, Max and Ultra; yearly-billed prices; export allowances; Batch exports count; separate API plans); remove.bg's pricing and API pages (a Canva brand; 50 free API calls a month); Canva's Background Remover help page (paid plans, 50-photo folders, 9 MB and 10 MP limits, no AI-content metadata); Google Merchant Center image requirements (AI metadata, size, framing, overlays); the author's own method. Check each tool's current terms, resolution limits and plan prices before batching. Marketplace image rules change: read your marketplace's current image page.