LayerGrab review: price, layers, limits and privacy

This page is built from LayerGrab's public pages, its Chrome Web Store, Figma Community and GitHub pages, the directories that list it, and its competitors' own pages, read on October 29, 2026. LayerGrab is our own product: it is made by the team behind GenGrades, and its splits run through apimodels.app, an AI API relay from the same team that we review separately. No independent test of LayerGrab has been published, so its numbers below are labelled as its own and its examples are called its own.

The short answer

LayerGrab takes one flat picture, such as a poster, a banner, a product photo or an AI image, and returns a stack of layers: each element on its own transparent layer with an English name, over a copy of the background with those elements painted out. You can ask for everything, up to 16 layers, or name only the parts you want.

It is paid per layer returned: $0.10 a layer on the $12-a-month Basic plan and the credit packs, $0.08 on the $36 Premium plan. The filled background counts as a layer, so a typical split of 7 to 10 layers plus the background costs about $0.80 to $1.10.

Choose it when you need pieces of an image you have no layered source for, as files or inside Figma. Skip it when the image is confidential, when you need editable text or vectors, or when the design is staying in Canva.

How a split works

On layergrab.com you drop in a PNG, JPG or WebP up to 20 MB, sign in with Google, optionally type what to separate, and wait: 60 to 100 seconds by its own figures, 46 to 148 seconds in its eight samples. You then hide layers to check what is behind them, copy one layer, or download PNGs, a ZIP or a layered PSD.

The same account works in a Figma plugin, approved on Figma Community on October 1, 2026, and a Chrome extension that splits any web image from the right-click menu. Photoshop and Sketch plugins are in testing. A canvas editor lets you move, delete or redraw layers and split one layer again into finer parts.

Each layer is redrawn by the model rather than cut out, so elements that were partly hidden come back whole, and very light, thin text can change colour slightly. Results are not repeatable to the layer: its homepage says one run can give 9 layers and the next 10.

A separate Qwen-Image-Layered page runs the open Qwen model on the same account: 2 to 10 unnamed layers at about 640 px, a different model from the main tool's.

What it costs in practice

The plans and packs on its pricing page, at 10 credits a layer (8 on Premium):

The per-layer figures for yearly plans and packs are our arithmetic. Subscription credits reset monthly; pack credits expire after a year. By the same arithmetic, one product out of a banner with its background is 2 layers, $0.20 on Basic, and a poster split into 10 layers and the background is $1.10, as its own worked example says.

Credits are held when a split starts and only delivered layers are charged; a failed split is refunded. If your balance cannot cover the result, LayerGrab locks it instead of charging part of it: you see each layer's name and position and can unlock all of them for the shown price within 30 days. Payment goes through Stripe, plans can be cancelled from the billing page, and the terms say credits have no cash value.

What the public evidence shows

The evidence about LayerGrab is almost all its own.

Its own examples. Its homepage shows eight real splits with every layer and its time: a coffee-shop poster in 11 layers, a headphones banner, a skincare post, two AI illustrations, a report slide, a sportswear flat-lay photo in 16 layers, and a two-person photo where 'Separate the woman on the right' returned one layer and the filled background. Six are published with their layers and positions in its samples repository. Nobody outside the company has published a repeat of them.

Listings. On October 29, 2026 the Chrome extension had 11 users and no ratings, the Figma plugin counted 11 uses and 2 likes, its GitHub repositories had no stars, and its Fazier launch had no upvotes. Directories such as HuntScreens and Findly repeat its own description. No Reddit, Hacker News or YouTube discussion and no independent review turned up in search.

So how clean the layers are on your kind of image has no outside answer yet.

Where it falls short

LayerGrab lists most of its weak spots itself, on its homepage and plugin pages:

• text comes back as a picture, not editable type, and shapes are not vectors;
• cast shadows stay in the background, so a moved element can look like it floats;
• the designer's fonts, masks and effects are not recovered;
• busy photos with overlapping objects split less cleanly than designed images;
• hair edges are softer than a careful manual mask;
• a split stops at 16 layers.

Its pages also disagree with each other:

• The pricing page promises one free split with every layer downloadable; the web tool's free button on October 29, 2026 extracted one named element as a single PNG.
• Its Sketch, Chrome and Qwen pages, Chrome Web Store listing and GitHub README still mention sign-up credits the pricing page does not offer.
• The Chrome and Figma pages say the plugins install in one click from their stores, but their step-by-step guides still show a developer-mode install.
• Its English homepage says it does not split characters into Live2D or Spine rigging parts; its LlamaGen and Japanese pages show such a split as a base for Live2D.

Trust the pricing page and what the tool shows after sign-in over the rest.

Privacy and data handling

Per its privacy policy, updated October 17, 2026, LayerGrab stores your Google email, name and picture, your uploads and the layers it returns. Results are deleted after 30 days; uploads stay with your account until you ask for deletion, which it does within 30 days of an email. Images go to "an AI model operated by our processing partner", which the policy does not name: it is apimodels.app, from the same team. The policy says images are used only for your results and not sold or used to train its own models, and it states no storage region.

The plugins and extension keep their connection key on your device, send only the image you pick and your instruction, and carry no advertising or analytics. The Qwen page sends images to Qwen-Image-Layered hosted on fal. The terms leave you the rights to your images and layers, and name no company, address, jurisdiction or governing law.

Who should use it and who should skip it

Use LayerGrab if:

• you need one element, or every element, out of a flat image whose layered source you do not have;
• you change headlines, prices or badges on finished posters and banners;
• you work in Figma and want the split dropped into your file;
• you want to move or remove one object in an AI image without generating it again.

Skip it if:

• the image is confidential or must not leave your machine;
• you need editable type, vectors or the original fonts and effects;
• the design is staying in Canva;
• you need a full Live2D or Spine rig in one pass;
• you only want one subject on transparency, where a background remover such as remove.bg is faster;
• you need a company with a named address and governing law, or a record of independent reviews.

LayerGrab vs Canva, Lovart, Qwen-Image-Layered, magiclayers.net and Photopea

LayerGrab's comparison pages are written by its makers, so these facts come from each product's own page. Canva's and Adobe's sites refused our page reader: Canva's limits are from its help page as search engines show it, Adobe's price from its plans page as read on October 25, 2026.

• Canva Magic Layers. Turns a flat image into movable elements inside the Canva editor. Canva's help page limits it to single-page PNG and JPG files up to 5,000 px on the longest edge, warns that photo-realistic images may not separate cleanly, and counts it against the plan's AI allowance. Choose it when the design stays in Canva.
• Lovart Edit Elements. Splits a selected image on Lovart's AI design canvas into layers, lets you change the font, size and wording of text elements, and exports a PSD, per Lovart's docs. Its pricing page lists a free plan with limited models and Starter at $16 a month billed yearly for 2,000 credits, without the cost of one split. Choose it if you design in Lovart and want text you can retype.
• Qwen-Image-Layered. The Qwen team's open model, under Apache 2.0, splits an image into the number of RGBA layers you choose; its README says the prompt is not designed to control what goes on each layer, and layers are unnamed. The official demo runs on Hugging Face's free daily GPU quota of 2 to 5 minutes (ZeroGPU tiers); fal charges $0.05 per image, as its page puts it, and Replicate $0.03 a run plus $0.01 an output image. Choose it when images must stay on your own GPU.
• magiclayers.net. You pick 2 to 15 output images, optionally describe what to separate, and get named transparent PNGs and a layered PSD, per its homepage. It charges 2 credits an image, with packs from a $2 trial to $69 for 1,000 credits (pricing): about $0.14 an image at best by our arithmetic, above LayerGrab's $0.08 to $0.11, but its $2 trial is a cheaper first purchase. Choose it to set the layer count yourself.
• Photopea or Photoshop, by hand. Photopea is a free, Photoshop-like editor that runs on your device with no uploads and full PSD support; Photoshop costs US$22.99 a month on an annual plan. You select each element and fill each hole yourself. Choose them for confidential images, a few elements or exact masks.