Generative texture signatures
Plastic skin, repeating hair and fabric patterns, foliage that dissolves into mush, and background text that is almost-but-not-quite letters.
Image forensics
AI image detector for generated and edited photos
Upload a photo and DetectFake checks whether it was fully AI-generated, locally edited by AI, or captured by a real camera — then shows the confidence score and every forensic signal behind the verdict. Free to start, nothing to install.
Analyze an image now 3 free scans every day
What the detector looks for
Each scan reports individual signals instead of one opaque score, so the reasoning is reviewable.
Local edit seams
Sharpness, grain and colour that change at the boundary of one object — the classic trace of an AI-added helmet, watch, person or logo.
Light and geometry errors
Shadows falling in disagreeing directions, missing contact shadows, reflections that omit the subject, and perspective that will not resolve.
Sensor and compression evidence
Noise that is too uniform for a real sensor, absent or resampled compression structure, and detail that survives where a camera would lose it.
How to check a photo in 4 steps
- 1
Upload the image
Drop a JPEG, PNG, WebP or HEIC file up to 25 MB. No install and no account needed for your first scans.
- 2
It is prepared in your browser
The photo is downscaled locally so analysis is fast and the full-resolution original stays on your device.
- 3
Forensic analysis runs
Every worn or added object, every lighting inconsistency and every texture anomaly is enumerated and weighted as a separate signal.
- 4
Read the verdict and evidence
You get a verdict, a confidence score, a plain-language summary, and the full signal list so you can defend the conclusion.
Limits worth knowing
Screenshots, social-platform re-compression, heavy filters and small thumbnails strip out the fine detail forensic analysis relies on, which pushes results toward inconclusive. Photos of screens, scans of prints and aggressive denoising can also mimic generative smoothness. Treat a result as strong evidence for triage — not as legal proof.
AI image detector FAQs
How accurate is an AI image detector?
No detector is exact. DetectFake returns a probabilistic verdict — Authentic, Manipulated, or AI-Generated — with a confidence score and the individual signals it found, so you can judge the evidence rather than trust a single number. Clean, high-resolution originals give the strongest results; heavily compressed screenshots give the weakest.
Can it detect Midjourney, DALL·E, Stable Diffusion and Sora output?
Yes. Rather than fingerprinting one specific model, DetectFake looks for the artifacts diffusion and transformer image models share: over-smooth or plastic texture, melted background text and signage, warped hands, teeth and jewellery, physically impossible reflections, and noise patterns that do not behave like a camera sensor.
Does it catch AI edits to a real photo?
That is the harder and more common case, and it is what DetectFake is calibrated for. Inpainting — an object added, a person removed, clothing or an accessory generated onto someone — leaves local seams, resolution and noise mismatches between the edited region and the rest of the frame, and lighting on the new object that disagrees with the scene. Any such signal caps the authenticity confidence.
Which image formats and sizes work?
JPEG, PNG, WebP and HEIC up to 25 MB. Images are downscaled in your browser before analysis, so large camera files are fine.
Is my photo uploaded or stored?
Only a downscaled copy is sent for analysis, and it is not retained after the report is produced. Your scan history is stored locally in your own browser.
How many images can I check for free?
Three scans per day on the free Regular plan. Silver raises this to 30 scans a day and Gold is unlimited with priority processing.
Check a photo for AI right now
Three free scans a day, no signup required.
Open DetectFake