Proof Logo
PROOF
Apple Download the App
Image Verification
- Results Issued
Upload Image
or, drag & drop an image here
preview
30+
Signals
|
5
Detectors
|
Beta
Version

How It Works
📸

Upload a Photo

5 physical stages trace how a camera sensor converts light into pixels.

🧬

Extract 30+ Signals

8 signal groups detect where fake images break real camera physics.

Deliver a Verdict

Built on a dataset of real devices to learn unique camera fingerprints.


Original Image
No manipulation detected
Example original camera image
Image Analysis · Strong
AI-Generated · 0%
Top 3 Decision Signals
1.Noise Spatial Variability 0%
Measures how much the grain pattern shifts between bright and dark areas of the image.
2.Noise Residual Strength 0%
Measures the faint grain a camera sensor leaves in every photo, like a hardware fingerprint.
3.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
Photo-of-Photo · 0%
Top 3 Decision Signals
1.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
2.Noise Spatial Variability 0%
Measures how much the grain pattern shifts between bright and dark areas of the image.
3.Gradient Entropy 0%
Measures how complex and varied the transitions are between bright and dark areas.
Screenshot · 0%
Top 3 Decision Signals
1.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
2.Noise Spatial Variability 0%
Measures how much the grain pattern shifts between bright and dark areas of the image.
3.Noise Residual Strength 0%
Measures the faint grain a camera sensor leaves in every photo, like a hardware fingerprint.
File Analysis · Strong
Web-Download · 0%
Top 3 Decision Signals
1.Platform Signature 0%
Checks for hidden marks that social media platforms embed when processing images.
2.EXIF Stripped 0%
Checks whether camera details like date, device, and location have been removed from the file.
3.AI-Typical Resolution 0%
Checks whether the image dimensions match known output sizes from AI generation tools.
Camera-Origin · 0%
Top 3 Decision Signals
1.Device ID 0%
Checks whether a device name is stamped in the file by the camera.
2.Date / Time 0%
Checks whether a capture timestamp is recorded in the file.
3.Native JPEG 0%
Checks whether the file structure matches how a camera saves photos versus software re-processing.
Image Details · Strong
FileIMG_4821.jpg
DeviceApple iPhone 15 Pro
DateJan 30, 2026 · 8:29 AM
Resolution1536x2048 · JPEG · 0.5 MB
Lensf/1.8 · 6.8mm
ExposureISO 250 · 1/60
Modified Image
Manipulation detected: AI-Generated
Example AI-generated image
Image Analysis · Weak
AI-Generated · 0%
Top 3 Decision Signals
1.Noise Spatial Variability 0%
Camera grain shifts naturally between bright and dark areas. Uniform grain across a scene suggests it wasn't optically captured.
2.Gradient Entropy 0%
Measures how complex and varied the transitions are between bright and dark areas.
3.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
Photo-of-Photo · 0%
Top 3 Decision Signals
1.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
2.Gradient Entropy 0%
Measures how complex and varied the transitions are between bright and dark areas.
3.Edge Density 0%
Measures the number of sharp, fine edges detected throughout the image.
Screenshot · 0%
Top 3 Decision Signals
1.Noise Residual Strength 0%
Measures the faint grain a camera sensor leaves in every photo, like a hardware fingerprint.
2.Microcontrast Strength 0%
Measures tiny surface textures and fine details visible at the pixel level.
3.Noise Spatial Variability 0%
Measures how much the grain pattern shifts between bright and dark areas of the image.
File Analysis · Weak
Web-Download · 0%
Top 3 Decision Signals
1.Platform Signature 0%
Checks for hidden marks that social media platforms embed when processing images.
2.EXIF Stripped 0%
Checks whether camera details like date, device, and location have been removed from the file.
3.AI-Typical Resolution 0%
Checks whether the image dimensions match known output sizes from AI generation tools.
Camera-Origin · 0%
Top 3 Decision Signals
1.Device ID 0%
Checks whether a device name is stamped in the file by the camera.
2.Date / Time 0%
Checks whether a capture timestamp is recorded in the file.
3.Native JPEG 0%
Checks whether the file structure matches how a camera saves photos versus software re-processing.
Image Details · Weak
Fileai_generated_selfie.png
Device
Date
Resolution1024x1792 · PNG · 2.1 MB
Lens
Exposure

Frequently Asked Questions

1. How does Proof work?

Every photo taken by a real camera passes through five physical stages. Each stage leaves traces in the pixels that are extremely difficult to replicate. Our 30+ signals measure each stage independently.

1. Photon capture
Shot noise from light hitting silicon. Extremely difficult to synthesize artificially.
2. Bayer demosaic
One color per pixel interpolated into three. Leaves a periodic pattern AI never produces.
3. Analog to digital
Each sensor burns a unique noise fingerprint (PRNU) into every pixel.
4. ISP processing
Apple, Samsung, Google, etc. each process images differently. Distinct artifacts in edges, gradients, and frequencies.
5. JPEG encoding
DCT blocks, quantization tables, and compression stats unique to each device.

If any stage is missing or altered, our signals detect it.

2. What does Proof detect?

Every image is scored by five independent detectors:

1. AI-Generated
Detects images created or edited by AI tools.
2. Photo-of-Photo
Detects photos taken of screens or prints.
3. Screenshot
Detects direct screen captures.
4. Web-Download
Detects images downloaded from web platforms.
5. Camera-Origin
Checks metadata provenance strength.

Each detector weighs the signals that matter most for that specific image and shows exactly which signals influenced its answer.

3. Can AI fool Proof?

AI generators synthesize pixels directly from neural networks. They skip the entire camera pipeline. No sensor noise, no Bayer pattern, no PRNU fingerprint, no hardware JPEG encoder. Our 30+ physics signals measure what should be there from a real camera. When it's missing, we know.

4. What makes Proof different?

Most AI detectors use neural networks trained to recognize what AI output "looks like." They break every time a new generator launches. Proof measures the physics fingerprint the camera pipeline leaves in every pixel. Real cameras produce signals that are extremely difficult to replicate, regardless of what AI tool was used. Every result shows exactly which signals drove the verdict. No black box.

5. Is my data stored?

All uploaded images are automatically deleted after 24 hours. Your data is never sold or shared. If you need an image removed sooner, contact contactus@proofme.ai.

View all FAQs →