FREE · PRIVATE · INSTANT
IS THIS IMAGE REAL OR AI-GENERATED?
Analyze photos locally in your browser. No upload, no account, no trace.
Drop image here
or click to browse · Ctrl+V to paste
UNCERTAIN
The synthetic image problem is here now
AI-generated imagery has reached photorealistic quality. Distinguishing real photographs from synthetic ones is no longer optional — it is a core information-literacy skill.
Fabricated images at scale
Synthetic images of events that never happened circulate on social media within minutes of a breaking news cycle. Fast, reliable first-pass detection is now essential for newsrooms and individuals alike.
AI-generated profile photos
AI headshots are used to create convincing fake professional profiles on LinkedIn, dating platforms, and in job applications. A plausible face is now free to generate in seconds.
Marketplace trust
Product listings, NFT collections, and stock photo libraries are increasingly polluted with AI-generated content passed off as original photography. Buyers deserve to know what they are purchasing.
Legal and academic integrity
Courts, insurance adjusters, and academic reviewers are encountering AI-generated imagery submitted as legitimate documentation, evidence, or portfolio work.
Who uses PixelTruth and why
Any workflow where image authenticity matters.
Journalists and Fact-Checkers
First-pass check on images submitted by sources or circulating on social media before publication. A high AI-probability score flags the image for deeper forensic investigation.
HR and Recruiting
Verify candidate profile photos before interviews. AI-generated headshots are widely used in social engineering and fraudulent applications.
E-Commerce and Buyers
Before purchasing art, photography, or digital products, verify whether images are authentic. Especially relevant for resale platforms and NFT marketplaces.
Educators and Academic Reviewers
Verify whether submitted portfolio work, photo assignments, or illustrated essays are genuine. Academic integrity policies are expanding to cover AI-generated visual content.
Personal Due Diligence
Before trusting a dating profile, a viral news image, or a social media post — drop it in. Get a second opinion in seconds. No account. No upload. No data shared.
How to spot an AI-generated image yourself
Tools help, but a trained human eye remains one of the most powerful detection instruments available.
Hands and fingers
The most reliable tell. Wrong finger count, fused digits, extra joints, impossible poses. Examine every hand in the frame, especially at the edges where models are weakest.
Text and signage
AI generators are trained on pixels, not language. Any text on signs, shirts, or labels will almost always be garbled. Read every word you can find in the image.
Ears and hair
Ears with no internal structure, hair that merges with objects, strands that disappear into complex backgrounds. Hair against windows or other people is especially revealing.
Perspective breaks
In real photographs all objects obey the same vanishing point. AI images frequently mix incompatible perspective geometry — walls, floors, and furniture with different angles.
Lighting inconsistency
Shadows falling in different directions on adjacent objects. Reflections in eyes or glasses that don't match the scene. Beautiful but internally contradictory lighting.
Jewelry and glasses
Glasses that don't align with ears, asymmetric earrings, necklaces floating off fabric. Small accessories are under-represented in training data and often rendered incorrectly.
Frequently asked questions
Can PixelTruth detect Midjourney, DALL·E, and Stable Diffusion images?
Yes. PixelTruth targets imagery from all major generators including Midjourney v5/v6, DALL·E 3, Stable Diffusion XL, and Flux. No detector achieves 100% accuracy as the adversarial landscape evolves rapidly, but our multi-signal fusion provides high-confidence results in most cases.
Is my image uploaded to a server?
No. The detection model is fetched once and cached by your browser. All processing — neural network inference and mathematical analysis — runs in a background Worker thread on your own device. We never see your images.
What does the confidence percentage mean?
The percentage is a fused signal from the neural network and five mathematical markers. 50% or above indicates AI-generated. Below 50% indicates a real photograph. The middle range (40–60%) is honest uncertainty — some images are genuinely ambiguous even for trained systems.
Why does the model take a moment to load the first time?
The detection model is approximately 52 MB and is downloaded directly from Hugging Face to your browser on first use. After that it is cached locally in IndexedDB — subsequent visits load it instantly without any network request.
Does JPEG compression affect accuracy?
Heavy JPEG compression (below quality 60) can reduce accuracy because it degrades the high-frequency signal patterns the model relies on. For best results use the original uncompressed file. Screenshots and social-media reposts are often already recompressed.