Where was this photo taken?

The AI location finder that geolocates any photo from pixels alone, with no GPS and no EXIF. Sub-second results, worldwide.

01

You upload

Any photo or screenshot. The file header is thrown away before analysis.

02

We read the picture

Architecture, terrain, vegetation and signage are weighed across the frame.

03

You get a place

Country, region, city, and the confidence radius that comes with it.

How it works

Three steps, one second

No metadata is read at any stage. The pipeline sees pixels, weighs visual cues, and returns a place with its confidence radius.

01

Pixels in, header out

The image is decoded and the file header discarded. EXIF and GPS fields are never parsed, so stripped or spoofed metadata changes nothing.

02

Cues weighed together

Architecture, terrain, vegetation, road markings and infrastructure are scored across the whole frame, the cues an analyst would name, at scale.

03

A place, and how sure

You get country, region and city, street level where the scene is distinctive, each with the confidence radius printed next to it.

The model

What it looks at in your photo

Five families of visual evidence, weighed at once. Nothing outside the frame is used, no web search, no database lookups.

Street scene annotated with the location cues the model reads
01 Architecture
02 Vegetation
03 Infrastructure
04 Terrain & light
05 Masonry & surfaces
01

Architecture

Render finish and colour, roof tiles, balustrade profile, window rhythm, here, flat-roofed white blocks with terracotta ridge tiles.

92%

02

Vegetation

Species mix and canopy state. Banana stands beside a paved lane narrow this to a subtropical, cultivated island slope.

86%

03

Infrastructure

Overhead cabling on concrete poles, kerb profile, the painted centre strip and cone spacing.

79%

04

Terrain & light

Downhill gradient towards a hazy sea horizon, overcast flat light, wet volcanic surface.

74%

05

Masonry & surfaces

Dry-stone basalt retaining wall with mortar joints and moss, a regional building method, not a global one.

68%

No facial recognition, ever.

Accuracy

The full accuracy curve, no cherry-picking

Image geolocation is probabilistic. We publish how the model performs at every distance threshold, not a single flattering number.

≤ 1 km

12.2%Geo-estimation
97%High-accuracy*

Street

The exact spot

≤ 25 km

61.20%

City

The right metro area

≤ 200 km

78.93%

Region

The right region

Country

93.3%

Country

The right country

Measured on the IM2GPS3k benchmark. Accuracy stays consistent across scene types; the model performs the same way whether the location is common or rare. * known areas.

Pixels only

No internet access required, no OCR, no label recognition, no agentic search. No EXIF, no web search, no database lookups. The prediction comes from the image itself, so it works on photos that were never posted online.

Soon

Agentic search

Coming as a separate, explicitly-labeled mode. It will cross-reference external sources for tighter results, so you always know whether a result came from pixels alone.

Capabilities

AI image geolocation, explained

Visual-only analysis that works when metadata is stripped, spoofed, or never existed.

No Metadata Required

Works when EXIF is stripped by social platforms, messaging apps, or manual removal. The model reads the image, not the file header.

Sub-Second Inference

Location predictions in under a second. Fast enough for live investigative workflows and real-time content review.

Global Coverage

Works in remote areas where Street View has no data. Urban cores, rural regions, conflict zones, anywhere on Earth.

Data Sovereignty

Your evidence never trains our public models. Zero-retention policies, AES-256 encryption, and full audit trails for chain-of-custody.

GPS-Denied Geolocation

Determine location from the image alone, in environments where GPS is unavailable, jammed, or blocked. No satellite signal required.

Batch & API Access

Process thousands of images in a single workflow. RESTful API for direct integration into investigation or compliance pipelines. Edge-native infrastructure scales horizontally without limits, no throughput ceiling under batch or high-frequency API load.

FAQ

Questions & answers

Common questions about GeoInfer and how image geolocation works.

GeoInfer geolocates images using AI, without GPS, EXIF, or any metadata. Upload a photo, and the model analyzes visual cues like architecture, terrain, vegetation, and infrastructure to determine where it was taken. Results in under a second, from anywhere in the world.

GPS-denied refers to environments where satellite positioning is unavailable, jammed, spoofed, or intentionally blocked. GeoInfer works from pixels alone, no GPS signal, no cell tower data, no metadata. This makes it useful in conflict zones, areas with signal jamming, and any situation where traditional positioning doesn't work.

Yes. GeoInfer doesn't use EXIF or GPS data. Social platforms and messaging apps strip location metadata automatically, and many images never had it to begin with. Our models analyze the visual content of the image itself, which is what makes this approach different from metadata-based tools.

We publish the full accuracy curve rather than a single number. The model is strongest at country and region level and lands the right city often but not always; exact-street accuracy is lower and depends on how visually distinctive the scene is. For investigative work, narrowing to the right region in under a second, then verifying, is the workflow that matters.

Under a second. The inference pipeline processes images in milliseconds, fast enough for live investigative workflows and high-volume batch processing. Results come back as soon as you upload.

Defense and intelligence agencies, law enforcement, insurance fraud investigators, investigative journalists, security analysts, and enterprise teams. Also used by visual geography enthusiasts, GeoGuessr players, and geography educators to identify location cues in images. Common professional uses: digital forensics, claims verification, source authentication, GPS-denied operations, and content provenance.

Yes. GeoInfer reads the same visual cues a skilled GeoGuessr player learns to spot, architecture style, road markings, vegetation type, terrain, and infrastructure. Upload a screenshot and the model returns a predicted location with a confidence radius. It's also useful for training: seeing which cues the model weighs most can sharpen your own pattern recognition for visual geography. The Chrome extension puts this inside GeoGuessr: press ⌘ Shift G (Ctrl Shift G on Windows) during a round for coordinates, a confidence radius and a mini map. Get the extension.

Yes. GeoInfer works with both ground-level and aerial imagery. Street scenes, building facades, and landscapes are geolocated using architectural and environmental cues. Aerial and UAV imagery is supported via the Pro tier, which handles overhead perspectives using terrain, land cover patterns, and infrastructure geometry.

We're on a SOC 2 certification path. Institutional users can opt for zero-retention policies, on-premise deployment, and air-gapped deployments, your data stays in your systems and never trains our public models. AES-256 encryption, full audit trails, and custom data handling agreements are available.

Try the demo in your browser — no account needed. Community API access is credit-based (top-ups start at $4.99). For Pro access with meter-level precision, on-premise deployment, or institutional licensing, contact our team at contact@geoinfer.com.

Start now

Drop one photo. See where it was taken.

It runs in the app with no account. Region-level answer in about a second, then decide.

  • Answer in about a second
  • No account for the demo
  • Zero-retention available
SEVILLE, ES37.39°N 5.98°WLISBON, PT38.70°N 9.14°WMARRAKESH, MA31.63°N 8.01°WLAGOS, NG6.52°N 3.38°ECAIRO, EG30.04°N 31.24°ENAIROBI, KE1.29°S 36.82°E~1 KM~25 KM~200 KM