Softonic ran a piece this week arguing the old giveaways for AI-generated images have stopped working. Six-fingered hands, glossy skin, garbled text on signs, wrong shadows. Nano Banana, Midjourney v7, and Stable Diffusion XL Turbo now clear most of them on the first pass. That leaves detection tools as the one honest reference for whether a viral photo actually came out of a camera. We tested the eight best apps for detecting AI-generated images on desktop, browser-based and native, against a sample of Nano Banana, DALL-E 3, and Midjourney v7 output.
Every pick here runs from a Windows, macOS, or Linux browser or client. None require a GPU. Most have a free tier that covers casual checks. We flag which ones bundle text and video detection too, because the same publications posting fake images are often also posting the article body around them.
What to look for in an AI image detector
- Model coverage across the current image generators. Midjourney, DALL-E, Stable Diffusion, Flux, Nano Banana, and Ideogram all leave different fingerprints. A tool that trained on 2023 models will miss 2026 output.
- Region highlighting. A confidence score alone is not enough. The best tools show which parts of the image triggered the detector, which helps for photo composites that mix real and synthetic content.
- Batch upload. Journalists and moderators check dozens of images at a time. Single-upload tools do not scale.
- API access for content teams. Newsrooms and platforms integrate the detector into their moderation pipeline. Free web-only tools cannot serve that use case.
- Honesty about limits. A vendor claiming 100 percent accuracy is a red flag. The strongest detectors publish their false-positive rate and update it as new generators land.
Quick comparison
| App | Best for | Free tier | Starting price | Standout feature |
|---|---|---|---|---|
| Hive AI Detector | Top overall accuracy across generators | Yes, browser demo | Custom API pricing | Trained on the most recent Midjourney and Nano Banana output |
| Illuminarty | Region heatmaps and generator attribution | 5 scans per day | From about $10 per month | Names the specific generator used |
| AI or Not | Simplest single-image check | Yes, limited scans | From about $5 per month | Fast binary verdict, no jargon |
| Sightengine | Newsroom and platform moderation | Trial credits | From about $29 per month | REST API, batch upload, video too |
| Winston AI Image Detector | Bundles text detection in one dashboard | Trial credits | From about $12 per month | Same login covers text, image, and code |
| Content Credentials Verify | Reading C2PA provenance metadata | Free | Free | Reads Adobe, Nikon, and Sony content credentials |
| Truepic Lens | End-to-end camera-to-editor provenance | Free tier | Enterprise pricing | Cryptographically signs images at capture |
| FotoForensics | Manual pixel-level forensic analysis | Free | Free | ELA, quantisation tables, metadata inspection |
The 8 best apps for detecting AI-generated images on desktop
1. Hive AI Detector, top overall accuracy
Hive AI Detector is the tool most independent 2026 benchmarks rank first. It trains against a rolling sample of the latest generators, including Nano Banana and Midjourney v7, and updates the model roughly every quarter. Upload an image on the web demo and you get a probability score, plus a “which generator” breakdown when confidence is high enough.
Where it falls short: The free web demo caps at low resolutions and rate-limits after a few uploads. Full accuracy is only available via the paid API.
Pricing:
- Free: web demo with limited daily uploads
- Paid: custom API pricing that scales with volume
Platforms: Web (any desktop browser), REST API for Windows, macOS, Linux clients.
Download: hivemoderation.com/ai-generated-content-detection
Bottom line: The pick when accuracy matters more than transparency, and you can justify API pricing.
2. Illuminarty, best region heatmaps and attribution
Illuminarty goes past a single confidence score. It shows a heatmap over the image marking which regions look synthetic, and tries to name the specific generator that produced them. That helps when the image is a mixed composite of real and AI content, which is now the more common case than fully synthetic images.
Where it falls short: The free tier caps at five scans per day. Attribution guesses on the newest generators are still tentative.
Pricing:
- Free: five image scans per day
- Paid: from about $10 per month for higher quotas
Platforms: Web, works from any desktop browser.
Download: illuminarty.ai
Bottom line: Best for image forensics work where you need to explain your reasoning, not just a verdict.
3. AI or Not, simplest binary check
AI or Not strips the interface to a single upload box and a two-word answer. That is exactly what most people want for the “did my friend send me a real photo” case. Turnaround is under two seconds for a typical JPEG. Their team also publishes clear accuracy numbers per generator, which is uncommon in the space.
Where it falls short: No heatmap, no attribution. Just AI or Not AI. That is a feature for one audience and a limitation for another.
Pricing:
- Free: limited daily scans without login
- Paid: from about $5 per month
Platforms: Web, plus a Chrome extension usable on Windows, macOS, and Linux Chrome.
Download: aiornot.com
Bottom line: The pick for a fast personal check on a shared image.
4. Sightengine, newsroom and platform moderation
Sightengine is the API most content-moderation teams plug into their pipeline. It covers image, video, and text in one contract, with per-request pricing. The web dashboard has a batch upload for spot-checks, and the AI detection model separately handles images generated by diffusion and by GAN architectures.
Where it falls short: Built for developers first. The web dashboard has power features but is not the fastest tool for casual, one-off checks.
Pricing:
- Free: trial credits without a card
- Paid: from about $29 per month with usage tiers
Platforms: REST API, dashboard runs from any desktop browser.
Download: sightengine.com
Bottom line: The pick if you are wiring detection into a moderation queue and want image, video, and text under one login.
5. Winston AI Image Detector, bundled text and image detection
Winston AI started as an AI text detector for educators and moved into image detection more recently. The strength is one login covering text, images, and code in the same dashboard, with a shared history of what you have checked. Their team publishes accuracy per generator and updates the numbers with each release.
Where it falls short: Image detection accuracy trails Hive and Illuminarty on the newest generators. Costs add up when you need volume across formats.
Pricing:
- Free: trial credits
- Paid: from about $12 per month for the standard plan
Platforms: Web dashboard, Chrome extension.
Download: gowinston.ai/ai-image-detector
Bottom line: The pick for teams that need text and image detection in one bundle, especially in an education setting.
6. Content Credentials Verify, reading C2PA metadata
Content Credentials Verify is Adobe’s free web tool for reading C2PA provenance metadata embedded in an image. If the file was signed at capture by a Nikon Z9, a Sony camera, or edited in Photoshop with credentials on, this tool decrypts the chain of custody and shows it. It is not a probability model, it is a metadata reader.
Where it falls short: Only works when the image carries C2PA metadata to begin with. That is a small share of what circulates on social feeds today, though it is growing.
Pricing: Free.
Platforms: Web, works from any desktop browser.
Download: contentcredentials.org/verify
Bottom line: The first check for any newsworthy photo. If credentials are present, they beat any probability model.
7. Truepic Lens, end-to-end signed provenance
Truepic Lens signs an image cryptographically at capture on supported devices, then lets viewers verify the whole chain up to the point they see it. It is the producer-side counterpart to Content Credentials Verify. On desktop, Truepic Vision reads and displays the signed chain, including camera, GPS, and any edits.
Where it falls short: Requires the source to have been captured through a Truepic-enabled workflow. Not a retrospective detector for existing images.
Pricing:
- Free tier for individual verifiers
- Paid: enterprise pricing for capture and signing infrastructure
Platforms: Web verifier, desktop reader on Windows and macOS, plus mobile capture apps.
Download: truepic.com
Bottom line: The pick for insurance, journalism, and evidence workflows where the photo needs a chain of custody, not just a confidence score.
8. FotoForensics, manual pixel-level analysis
FotoForensics is the old-school forensics tool that predates the diffusion era, and it still earns a place on this list. It runs Error Level Analysis, exposes quantisation tables, reads EXIF, and shows a colour histogram. On synthetic images, ELA often reveals a suspicious uniform noise pattern the diffusion models cannot help but leave.
Where it falls short: Requires a human interpreter. The tool shows the raw evidence, not a verdict. A visual-forensics background helps.
Pricing: Free.
Platforms: Web, works from any desktop browser.
Download: fotoforensics.com
Bottom line: The pick for anyone who wants to look at the file themselves rather than trust a black-box model.
How to pick the right one
- If you want the highest accuracy on the newest generators: Hive AI Detector.
- If you need to show your reasoning: Illuminarty for regional heatmaps, or FotoForensics for the raw pixel evidence.
- If you just want a fast yes or no on a photo a friend sent: AI or Not.
- If you run a moderation queue: Sightengine for the API, Winston AI for a bundled dashboard.
- If a photo is going to matter in court, insurance, or news: Content Credentials Verify first, and Truepic Lens for the producer side of the chain.
Skip any tool that will not publish its accuracy per generator. Skip any that claims 100 percent.
FAQ
What is the most accurate AI image detector in 2026? Hive AI Detector leads most independent 2026 benchmarks across Midjourney, DALL-E, and Nano Banana output. Illuminarty is close on Midjourney and Stable Diffusion specifically.
Is there a free AI image detector? Yes. AI or Not, Illuminarty, Content Credentials Verify, and FotoForensics all have functional free tiers. Hive and Sightengine offer free demos with limited quotas.
Can these tools detect deepfakes? Some can. Sightengine and Winston AI cover video deepfakes as well as still images. Deepware Scanner focuses specifically on video deepfakes, though it now runs as a web tool rather than a mobile app.
Do AI detectors work on edited or filtered photos? Detection accuracy drops on photos that have been compressed, cropped, or filtered after generation. Hive and Illuminarty publish their numbers on post-processed images; both are lower than on raw output.
What is C2PA and why does it matter? C2PA is a content-authenticity standard that lets cameras and editors cryptographically sign an image and its edit history. If a photo carries C2PA credentials, provenance beats any probability model. Content Credentials Verify is the easy way to read them.
