Turn phone photos into marketplace-ready listings

Yes. There’s no in-browser cloud demo — instead, the full app is free for 7 days on your own computer, every feature unlocked, no credit card, just your email. That means you test it at full speed on your actual hardware before paying a cent, so it’s never a blind purchase. Buy a license anytime after to keep going, and there’s a 45-day money-back guarantee on top of that.

Yes. $49 and it's yours permanently. No subscriptions, no credits, no usage limits. Process one image or one million — same price. All future updates are included at no extra cost.

The other tools run AI on their cloud GPUs, so every photo you process costs them server time — that's why they charge $13–35/month or per-credit. ListingGems runs on your own computer, so processing has zero marginal cost on our end. That cost structure is what makes a $49 one-time price sustainable.

It's also a solo-founder venture, not a VC-backed company — no investor pressure to maximize revenue per user means no subscriptions, no usage caps, no upsells.

Three models to choose from:

• BiRefNet full (~470 MB) — highest quality
• BEN2 (~210 MB) — balanced
• BiRefNet lite (~110 MB) — fastest

Both BiRefNet (2024, Nankai University and collaborators) and BEN2 (2025, Prama LLC) report state-of-the-art results on the DIS-VD benchmark — the standard test for clean cutouts on fine details like hair, fur, and translucent edges. BEN’s 2025 paper edges out BiRefNet by a small margin (S-measure 0.917 vs 0.905). For typical marketplace products the segmentation quality matches what you’d get from cloud tools like remove.bg or Photoroom.

Model cards: BiRefNet ONNX · BEN2

Very close. On the BiRefNet authors’ own DIS-VD benchmarks, the lite model scores about 95–97% of the full model on the main segmentation metrics (S-measure 0.882 vs 0.911, weighted F-measure 0.830 vs 0.875). In practice on typical marketplace products — clothes, shoes, electronics, jewelry, homewares — the difference is hard to spot, which is why we ship lite as an option.

The full model has a noticeable edge on the very hardest cases: long flyaway hair, fine lace, semi-transparent fabrics, complex fur patterns. Pick full if you do a lot of those, or BEN2 if you want balanced quality with a smaller download than the full BiRefNet.

Only for the initial model download (about 110–470 MB depending on the model you choose). After that, ListingGems works completely offline. No cloud, no API calls, no data sent anywhere.

No. ListingGems runs on any modern Windows or Linux computer — no graphics card required. On CPU it takes about 20–30 seconds per image, which is fine for everyday listing work. If you happen to have a GPU (NVIDIA GTX 1650 / 2019 or newer with 4GB+ VRAM, or an AMD GPU on Windows), processing drops to under 2 seconds per image — but it’s a speed bonus, not a requirement.

The honest answer: you need a computer. ListingGems is a desktop app for Windows and Linux (macOS coming) — you can’t run it from your phone or tablet. That’s the trade-off for keeping your photos on your own machine and charging a one-time price instead of a subscription. If your whole selling workflow lives on a phone, a mobile app like remove.bg or Photoroom is the better fit. Two smaller things to know:

• No team / cloud library. Every install is independent. A 5-person agency would need 5 separate licenses (or use the 3-device activation across a small team). For solo sellers and small operations this isn’t a blocker; larger teams that want a shared cloud workflow may prefer a SaaS tool.
• Very large batches on a slow PC. It runs on any computer, but without a GPU each photo takes 20–30 seconds instead of ~2. Fine for everyday listing; if you process hundreds of photos in one sitting on an older machine, expect it to take a while (you can leave it running — it’s all offline).

Amazon (Main + Secondary), Etsy, eBay, Poshmark, Mercari, Depop, Facebook Marketplace, and Shopify. Each preset auto-crops to the platform's required dimensions. Amazon Main enforces a pure white background.

No — every major resale marketplace (Depop, Poshmark, Mercari, eBay, Vinted, Facebook Marketplace) bans stock photos, even for genuine items. The brand owns the photo under copyright law, and platforms enforce this with permanent suspensions and frozen balances. Use a phone photo of the actual item you're selling. Full breakdown by platform →

Input: JPEG, PNG, WebP, HEIC, and AVIF — drop photos straight from your iPhone or Galaxy, no conversion needed. Output: JPEG, PNG (with transparency), and WebP — with marketplace-specific optimization for each format.

Yes. The AI models handle complex edges including glass, transparent packaging, jewelry, and reflective surfaces. For tricky cases, the built-in touch-up brush lets you refine edges by hand.

Generative AI tools invent pixels. They're built to generate plausible imagery — which means they smooth textures, idealize surfaces, and hallucinate details that aren't on your actual product. That's fine for lifestyle scenes (your product in a kitchen, on a beach, in a model's hand). It's risky for Amazon main images, where the policy requires the photo to accurately represent the real product. Misrepresentation gets listings pulled.

ListingGems uses ML segmentation, not generation. It identifies where your product ends and the background begins, then swaps the background. Your real photo of your real product — just background-cleaned and Amazon-sized. No invented details, no policy risk.

Use generative AI for lifestyle shots. Use ListingGems for the photo Amazon actually grades you on. Full Photoroom comparison →

ListingGems includes a manual eraser and restore brush for edge refinement. If the AI misses a spot or removes too much, you can paint it back in or erase it with precision.

Your images never leave your computer. There is no cloud connection, no upload, no analytics on your photos. The AI models run entirely on your hardware. Ideal for sellers working with unreleased products, private label designs, or confidential supplier materials.

Full refund within 45 days, no questions asked. Email support@listinggems.com with your order number or purchase email, and we'll process it within 48 hours. You can also go through our payment provider Creem if you prefer — full details on the refund page.

Checkout is processed by Creem (Armitage Labs OÜ, registered in Estonia), an EU-based payment provider for software companies. They handle the card form, PCI compliance, and VAT — ListingGems never sees or stores your card details. When you click Buy Now, you'll be taken to a Creem-hosted checkout page; that's expected.

Yes — each license supports up to 3 device activations. You can deactivate a device anytime from Settings to free up a slot.

macOS support is coming soon. The ML infrastructure for Apple Silicon (CoreML) is already built in — we just need to test and package it.

Upscaling uses HFA2k, a neural network that enhances detail and sharpness. Speed depends on image size and your hardware. With a GPU: about 1 second per tile (~1 minute for a 2048×2048 image). On CPU: about 8 seconds per tile. Images larger than ~2048×2048 pixels are too large to upscale — crop or resize them first. Background removal and lighting correction are separate steps and remain fast regardless of upscaling.