Labelme
Offline AI image annotation for YOLO and VOC datasets — local SAM2, SAM3 and YOLO-World, one-time purchase from $49.
If your images can't leave the machine, Labelme is the most practical annotation app you can buy: local SAM2, SAM3 and YOLO-World prompting at $49 Starter or $79 Pro one-time, with a 14-day no-questions refund and a 3-day full-app trial. The Pro toolkit is what actually makes it a dataset tool — batch box/text annotation, YOLO, YOLO-OBB and VOC export, import from YOLO and YOLO-OBB — and it runs in the terminal with Python rather than in the GUI. What you are not buying is a collaboration platform: there's no hosted workspace, review queue, or browser UI, so distributed labeling teams should look at Roboflow or Supervisely instead. Solo practitioners and small regulated teams should start
Verified 5d ago · liveness 69/100 · cite: rightaichoice.com/tools/labelme
- Computer vision engineers building private YOLO or VOC training datasets
- Medical, satellite and industrial imaging teams that cannot upload source data
- Solo practitioners who want a permanent tool instead of a subscription
- Air-gapped or firewall-restricted environments that can use the Labelme model mirror
- Distributed labeling teams needing real-time multi-annotator collaboration or a review queue
- Buyers who want a browser-based or mobile annotation tool
- Projects that need a documented programmatic API for automated annotation pipelines
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Labelme if you need multiple annotators working in a shared browser workspace with a review queue and role permissions — Labelme is a single-machine desktop app with no documented cloud collaboration or API layer.
Your machine has to run SAM2, SAM3 and YOLO-World locally, so a GPU-less laptop that manages annotation today may feel slow once you lean on box and text prompts.
Starter is $49 one-time and Pro is $79 one-time, with Pro (Lifetime) at $249 one-time; Team licenses are arranged by email with volume pricing. That sits far below monthly cloud annotators like Roboflow or Supervisely, where a small team can spend $49-$79 in two or three months and lose access when billing stops. It's above the $0 open-source pip install, which is the same annotation app with no packaging, no mirror downloads and no support.
In short
Labelme — Offline AI image annotation for YOLO and VOC datasets — local SAM2, SAM3 and YOLO-World, one-time purchase from $49. Best for Computer vision engineers building private YOLO or VOC training datasets, Medical, satellite and industrial imaging teams that cannot upload source data, Solo practitioners who want a permanent tool instead of a subscription. Free to start; paid plans from $49.
What's new in Labelme
Checked 5 days agoAcross the latest 5 updates: 5 feature updates.
Set Shape Flag defaults in Settings
v7.6.0 lets you match Labels with patterns and define the Shape Flags each matching Shape should offer, without editing the Config File.
Download and manage AI models in Settings
Prepare AI models before an annotation session, keep annotating while they download, and cancel or delete them from Settings.
Move between neighboring Shapes without the mouse
Ctrl+Arrow, or Command+Arrow on macOS, selects a visible neighboring Shape while the Shape List stays in sync.
Search Settings by name and jump straight to the control
Search Settings by a familiar name or related term, then open the matching control without knowing which section contains it.
Mask fragments become one mask with one Undo to restore them
v7.5.0 merges mask fragments that share a label into one annotation, keeping every marked pixel; one Undo brings the original fragments back.
Viability Score
How well maintained and how widely used is Labelme? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Offline AI annotation with local SAM2, SAM3 and YOLO-World models
- SAM2 point prompts — click to include, shift-click to exclude, returns an editable polygon
- SAM3 box prompts — one rough box returns oriented boxes on every instance
- SAM3 text prompts — type a phrase like "green plant" and matching objects are annotated
- Live polygon detail adjustment while the AI preview is still on screen (v7.2.0)
- Shapes: polygon, mask, rectangle, oriented rectangle, circle, line, point
- Multi-gigapixel and float32 GeoTIFF support for satellite and medical imagery
- Pro toolkit: batch AI annotation with box or text prompts
- Export to YOLO, YOLO-OBB and Pascal VOC; import from YOLO and YOLO-OBB
- Mask and visualization generation from annotations; dataset statistics and label lists
- Rename labels across a dataset and resize images with their annotations
- Automatic, Uniform or By Label shape color schemes (v7.2.0)
- Cursor-anchored Ctrl/Cmd-wheel zoom (v7.2.0)
- Undo immediately after the first shape, with auto-save (v7.2.0)
- Download and manage AI models from Settings, including cancel and delete (v7.6.0)
About Labelme
Labelme is a desktop image annotation app that turns a local folder of images into a training-ready dataset without uploading anything. AI models — SAM2 for point prompts, SAM3 for box and text prompts, and YOLO-World — download once through the in-app Settings (added in v7.6.0, September 2026) or via a Labelme mirror that works behind corporate firewalls, then run entirely on your machine. No API key, no third-party AI account, no GPU rental. Because nothing is uploaded, it suits computer vision engineers and teams in medical, satellite, agriculture, and industrial imaging who cannot send source data to a cloud service. On top of open-source Labelme (16K+ GitHub stars, maintained since 2016), it adds the standard shape set — polygons, masks, rectangles, oriented rectangles, circles, lines, points — plus multi-gigapixel and float32 GeoTIFF support, dark mode, and an interface in 14 languages. The Pro tier ($79 one-time) unlocks a terminal-based dataset toolkit with batch AI annotation, export to YOLO, YOLO-OBB and Pascal VOC, import from YOLO and YOLO-OBB, mask/visualization generators, dataset statistics, label renaming, and resizing. Starter and Pro each include one year of app updates; the version you have keeps working afterward, with no license check in the app. Against cloud platforms like Roboflow or Supervisely, Labelme trades collaboration and hosted APIs for data sovereignty and a one-time price.
Behind the Verdict
Labelme's pitch is narrow and honest: your images never leave your computer, and you pay once. That combination is rare enough that it's the whole reason to buy. The AI layer is what makes offline annotation tolerable rather than painful. SAM2 handles point prompts — click to include, shift-click to exclude, and you get a polygon back that stays editable point by point. SAM3 handles box prompts (draw one rough box on one car, get oriented boxes on every car) and text prompts (type "green plant" and the orchard survey gets annotated). YOLO-World rounds out the set. Model management moved into Settings in v7.6.0 (September 2026), so you can prepare models before a session, keep annotating while they download, and cancel or delete them without leaving the app. For air-gapped machines, the docs publish a SAM3 offline pack with SHA-256 checksums you transfer by hand. The non-AI parts have been getting real polish. v7.2.0 added live polygon detail adjustment while an AI preview is still on screen, Automatic/Uniform/By Label color schemes, cursor-anchored zoom, and undo immediately after your first shape on a raw image. v7.5.0 merges mask fragments that share a label into one annotation with a single Undo to restore them. v7.6.0 added searchable Settings, Ctrl+Arrow/Command+Arrow navigation between neighboring shapes, and Shape Flag defaults configurable without editing the config file. The Windows installer and portable exe have been digitally signed by Kentaro Wada since v7.4.1. The audit trail matters for regulated buyers: shapes include polygons, masks, rectangles, oriented rectangles, circles, lines and points, with multi-gigapixel and float32 GeoTIFF support for satellite and pathology work. There is a 32,767-pixel limit on either side, and very large images ask for confirmation before opening — a memory safety limit, not a license limit. The limits are structural, not fixable with a patch. The Pro toolkit runs in the terminal and requires Python, which is a strange fit for a product whose Starter pitch is "no Python required." Starter to Pro is a separate purchase with no credit for what you already paid, so buying Starter and upgrading later costs more than going straight to Pro. There's no browser UI, no mobile app, and no documented programmatic API for automated pipelines. And the AI models are only as good as the prompts you write — box and text prompting cuts clicks, it doesn't remove review. Where it fits: solo practitioners, small teams, and anyone whose data governance rules out cloud upload, annotating for custom YOLO or VOC training. Where it doesn't: distributed labeling operations, browser-first workflows, or teams that need an API to slot annotation into an existing pipeline.
Researching Labelme? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Labelme actually fits — and what changes day-one when you adopt it.
Install the Starter desktop app, run the 3-day trial, open a folder of 2,000 road images, and use SAM2 point prompts plus SAM3 box prompts to outline vehicles.
Outcome: An editable polygon set per image that stays on disk, with auto-save on and undo available from the very first shape.
Buy Pro, transfer the SAM3 offline pack and its SHA-256 checksum to the air-gapped workstation, extract the weights into the osam cache, and annotate pathology slides with float32 GeoTIFF support.
Outcome: A labeled dataset that never touched a cloud service, exported from the terminal toolkit to Pascal VOC for training.
Open large aerial frames (confirming the memory-safety prompt when it appears), use SAM3 text prompts to mark orchards, then merge mask fragments that share a label with one Undo.
Outcome: A cleaned oriented-box dataset exported to YOLO-OBB without leaving the workstation.
Use Cases
- Annotate thousands of images offline with AI assistance for custom YOLO training.
- Create oriented bounding box datasets for YOLO-OBB training using the Pro toolkit.
- Prepare private medical or satellite imagery datasets without uploading source data.
- Batch-convert and export existing annotations to YOLO, YOLO-OBB or Pascal VOC.
- Use SAM3 text prompts to auto-annotate repeated objects across a survey folder.
- Merge mask fragments into single annotations before export, with one Undo if you change your mind.
Models Under the Hood
as of 2026-09-23
Limitations
- Labelme is offline-first: the AI models download once and then run on your machine, so images, annotations and exports never leave your computer, and no third-party AI account or API key is needed.
- One consequence is that no cloud collaboration or API-based workflow is documented — there is no hosted workspace and no programmatic annotation API.
- Model management (download, cancel, delete) happens in the in-app Settings, and weights normally come through the Labelme mirror rather than GitHub or Hugging Face.
- There is a 32,767-pixel limit on either side of an image; very large images that hit the memory safety limit can open only after you confirm.
- Current macOS builds require Apple Silicon — Intel Macs are supported only up to v7.0.4.
- The Pro dataset toolkit runs in the terminal and requires Python, so the "no Python required" promise applies to the annotation app, not the export tools.
- Windows may show a SmartScreen warning even on the signed v7.4.1+ releases.
as of 2026-10-03
Verification history
We have re-verified Labelme 7 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 7 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Labelme tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free Open Source
$0
Ideal for
Engineers comfortable with pip and dependency management who want every annotation feature and the local AI models at no cost.
What this tier adds
Free entry point — the GPL-3.0 app from PyPI or a Linux package; you handle Python setup, GitHub/Hugging Face weight downloads, updates and community support yourself.
Starter
$49 USD one-time
Ideal for
Solo annotators and small teams who want the tested desktop build without touching Python, including users behind a corporate firewall.
What this tier adds
Adds the ready-to-install macOS/Windows/Linux app with one-click SAM2 and SAM3 text annotation, model downloads via the Labelme mirror, and 1 year of app updates.
Pro
$79 USD one-time
Ideal for
Anyone actually producing training datasets — YOLO, YOLO-OBB or VOC — rather than just annotating images one at a time.
What this tier adds
Adds the terminal dataset toolkit: batch box/text annotation, YOLO/YOLO-OBB/VOC export, YOLO and YOLO-OBB import, mask and visualization generation, dataset statistics, label renaming, image resizing, and priority email support at 48h.
Pro (Lifetime)
$249 USD one-time
Ideal for
Long-horizon projects that will still be annotating in three-plus years and don't want to re-buy an update window.
What this tier adds
Everything in Pro plus every future update, including major versions, for one payment — priced near three years of Pro renewals.
Team
Custom
Ideal for
Small organizations that need multiple seats, consolidated billing, or procurement-friendly invoicing.
What this tier adds
Every seat gets Pro on one invoice with volume pricing, arranged by email with a 1-business-day response.
Where the pricing makes sense
The company stage and team size where Labelme's pricing actually pencils out — and where peers do it cheaper.
Starter is $49 one-time and Pro is $79 one-time, with Pro (Lifetime) at $249 one-time; Team licenses are arranged by email with volume pricing. That sits far below monthly cloud annotators like Roboflow or Supervisely, where a small team can spend $49-$79 in two or three months and lose access when billing stops. It's above the $0 open-source pip install, which is the same annotation app with no packaging, no mirror downloads and no support.
Setup time & first value
How long it actually takes to get something useful out of Labelme — broken out by persona, not the marketing-page minute.
Starter and Pro: download one installer (Windows installer or portable exe, macOS Apple Silicon .dmg, or Linux AppImage) and you can annotate within minutes; the first AI run is delayed only by the one-time model download, which you can now queue from Settings while you work. Pro toolkit: expect a longer first session — the export tools run in the terminal and require a working Python
Switching to or from Labelme
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From the open-source pip install: buy Starter or Pro and open the same folder — annotations and the ~/.labelmerc config carry over, and the paid build skips dependency wrangling.
- →From Roboflow or Supervisely: export your existing annotations as YOLO or Pascal VOC, then import them with Labelme's import-from-yolo or import-from-voc toolkit commands.
- →From CVAT or Label Studio: export to Pascal VOC or YOLO and bring the files into a Labelme folder for local editing.
- →From Labelme v7.0.4 on an Intel Mac: migrate to an Apple Silicon machine, since current builds require it.
- ↗To Roboflow or Supervisely: export YOLO, YOLO-OBB or Pascal VOC from the Pro toolkit and upload to the hosted platform when you need multi-annotator workflows.
- ↗To a custom PyTorch pipeline: use the documented Python readers for Labelme JSON and the export-to-* toolkit commands to feed training code directly.
- ↗To Ultralytics YOLO training: export YOLO or YOLO-OBB and point your dataset config at the output folder.
Resources & Guides
- Documentationlabelme.io
Docs · Labelme
Full product docs from labelme.io
- Documentationlabelme.io
Troubleshoot · Labelme
Full product docs from labelme.io
- Documentationlabelme.io
Ai Prompts And Models · Labelme
Full product docs from labelme.io
- Documentationlabelme.io
Dataset Guide · Labelme
Full product docs from labelme.io
- Resourcelabelme.io
Blog · Labelme
Helpful link from labelme.io
Tutorials & Learning
YouTube returned 6 videos for “Labelme”, and we withheld 6: 6 could not be judged, because “Labelme” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Labelme.
Official links
Tools that pair well with Labelme
Common stack mates teams adopt alongside Labelme, with the specific reason each pairing earns its keep.
Cvat
Open-source, multimodal data annotation platform for image, video, 3D point cloud, and audio labeling—self-hosted, cloud, or fully managed
Markov
Human-recorded computer-use datasets that teach AI agents to operate real software the way people do.
Fashion Mnist
A free 70,000-image Zalando clothing dataset built as a drop-in replacement for MNIST.
Featured Head-to-Head Comparisons
Labelme vs The New Black
For fashion professionals accelerating design-to-prototype, The New Black is purpose-built with specialized AI features like tech pack export and virtual try-on. For computer vision teams needing private, offline annotation, Labelme offers unmatched control with on-device AI and no recurring fees. Choose based on your domain: fashion creation vs. dataset preparation — they serve fundamentally different needs.
Labelme vs Adobe Firefly Services
Choose Adobe Firefly Services if you need generative image APIs at scale with enterprise compliance and Adobe ecosystem integration. Choose Labelme if you require offline, AI-assisted annotation for computer vision datasets, especially for YOLO training, with data sovereignty.
Labelme vs Qoves
If you want a science-backed, non-surgical glow-up plan based on your facial biometrics, QOVES is your choice—it's a one-time analysis with detailed recommendations. If you need a private, offline image annotation tool for computer vision datasets with AI assistance, Labelme is the clear winner—it offers a free open-source version plus affordable one-time paid options. These tools serve entirely different purposes, so your decision depends on whether you're improving your face or labeling images.
Alternatives to Labelme
View all
Cvat
Open-source, multimodal data annotation platform for image, video, 3D point cloud, and audio labeling—self-hosted, cloud, or fully managed
Markov
Human-recorded computer-use datasets that teach AI agents to operate real software the way people do.
Fashion Mnist
A free 70,000-image Zalando clothing dataset built as a drop-in replacement for MNIST.
Frequently Asked Questions
Best-of guides
Used Labelme? Help shape our editorial sentiment research.