How to Tell If a Photo Is Photoshopped — 7 Detection Methods
Someone sends you a photo and something feels off. The lighting is too perfect, the background doesn't quite match, or a detail just looks wrong. Here's how to move from gut feeling to actual evidence.
The tricky part about Photoshop edits is that the good ones are invisible. An amateur clone-stamp job leaves smeared textures and repeated patterns you can spot at full zoom. But a skilled retoucher blends edits so seamlessly that no amount of squinting at the screen will help. That's where forensic analysis comes in — tools that examine the file's internal structure, compression artifacts, and metadata to find traces the editor couldn't erase.
Try it free: Authenticity Checker — Detect Photoshop editing and manipulation. Runs in your browser, no signup needed.
None of these methods are perfect on their own. Each one catches a different type of edit. The real power comes from layering multiple checks: if two or three independent methods flag the same region of an image, you've probably found something.
1. Check the Metadata First
Every digital photo carries embedded metadata — camera model, lens, shutter speed, GPS coordinates, timestamps, and crucially, a Software field. When someone opens a photo in Photoshop and saves it, Adobe writes its name directly into the file: "Adobe Photoshop 25.3" or "Adobe Photoshop Lightroom Classic 13.1." This is the lowest-effort check you can do, and it catches a surprising number of edits because most people don't think to strip metadata after saving.
Upload the photo to the EXIF Checker and look for the Software tag. If it says Photoshop, Lightroom, GIMP, Pixelmator, or Affinity Photo — the file was processed through editing software. That alone doesn't prove manipulation (photographers legitimately adjust exposure and color), but it establishes that the image is not straight from a camera.
Also check the timestamps. The DateTimeOriginal field records when the shutter fired. The ModifyDate field records the last save. A gap of hours or days between them means the file was reopened and re-saved — possibly after edits. The EXIF Compare tool is useful here if you have a second photo from the same camera session to cross-reference settings.
2. Error Level Analysis (ELA)
ELA works by re-saving the image at a known JPEG quality level and comparing the result to the original. Areas that were edited after the initial save will compress differently than the untouched background. These differences show up as bright spots on a heatmap.
The idea is straightforward: a photo taken by a camera and saved once has uniform compression across the entire frame. If someone pastes in an element from a different image — a person, an object, a background — that pasted region was compressed at a different quality level or not compressed at all. When ELA re-compresses the image, the pasted region reacts differently than the surrounding pixels.
Upload a suspicious image to the ELA Scanner, and look at the heatmap. Uniform brightness across the image suggests no tampering. A bright region against a dark background (or vice versa) points to a section that was modified. Adjust the sensitivity slider to fine-tune what shows up.
One caveat: ELA has false positives. Sharp edges, high-contrast areas, and solid-color regions naturally produce different error levels. A red stop sign against a blue sky will look "different" even in an unedited photo. ELA is a flag, not a verdict — use it alongside other methods.
Upload a photo to run automated metadata, compression, and forensic analysis — detect edits in seconds.
Check Photo Authenticity →3. Clone Detection
Clone stamping is the most common Photoshop operation for removing objects. The retoucher paints over an unwanted element — a person, a sign, a blemish — by copying texture from a nearby area. The result looks natural because the replacement texture matches the surrounding context. But it creates duplicated pixel patterns that don't occur naturally in photographs.
The Clone Detection Scanner finds these duplicated regions using DCT fingerprinting. It divides the image into small blocks, computes a signature for each one, and searches for blocks with near-identical signatures that are far apart spatially. When it finds matches, it highlights them with colored overlays and draws connection lines between the duplicated areas.
This method is particularly effective for detecting removed objects and duplicated backgrounds. If someone cloned a patch of grass to cover a person who was standing on a lawn, the scanner will find two areas of grass with suspiciously identical texture patterns.
4. JPEG Ghost Analysis
When a JPEG is saved, the encoder compresses it at a specific quality level — say, quality 85. If someone opens that JPEG in Photoshop, pastes in a region from a different image (which was saved at quality 72), and re-saves the whole thing at quality 90, the pasted region now has a different compression history than the rest of the image. It was compressed twice: once at 72, once at 90. The original background was compressed twice too: once at 85, once at 90. These different histories leave different artifacts.
The JPEG Ghost Scanner sweeps through quality levels from 60 to 99, re-compressing at each level and measuring how much each pixel changes. At the quality level matching the original save, the untouched areas barely change (they're already at that quality). At the quality level matching the pasted region's original save, that region barely changes. The mismatch between these responses shows up as a "ghost" — a region that behaves differently from its surroundings.
This is one of the most powerful detection methods for composite images where elements from different sources were combined. It catches edits that ELA might miss, particularly when the pasted element has similar visual content to its surroundings.
5. Noise Consistency Analysis
Every camera sensor produces a characteristic noise pattern — random pixel-level variations that are consistent across the entire frame. When you paste in a region from a different camera, a different ISO setting, or a synthetic source, the noise pattern breaks. The pasted area will have different noise characteristics than the rest of the image.
The Noise Analysis Scanner extracts the noise residual by subtracting a median-filtered version of the image from the original. The remaining noise is then analyzed for consistency. Regions with anomalous noise levels — significantly smoother or rougher than their neighbors — get flagged.
This method catches edits that blur tools can't hide. An editor might blend the edges of a pasted region perfectly, but they can't easily match the noise floor. Professional forensic examiners consider noise analysis one of the most reliable indicators of image manipulation.
6. Shadow and Lighting Checks
Composited photos often get the lighting wrong. If someone pastes a person into a scene, the person's shadows might fall at a different angle than the shadows of objects already in the scene. The color temperature of the light on the pasted subject might not match the ambient light. These inconsistencies are sometimes visible to the eye, but not always.
The Shadow Consistency tool analyzes shadow directions across the image. The Luminance Gradient Scanner maps the brightness surface to reveal lighting direction. If you see a gradient flowing left-to-right in the background but top-to-bottom on an inserted object, that's a strong sign of compositing.
7. Thumbnail Mismatch
JPEG files embed a tiny thumbnail image in their EXIF data. Cameras generate this thumbnail at the moment the photo is taken. When someone edits the main image in Photoshop but doesn't regenerate the thumbnail, the thumbnail still shows the original, unedited version. Comparing the two reveals exactly what changed.
The Thumbnail Scanner extracts the embedded thumbnail and displays it alongside the main image. If the thumbnail shows a person who isn't in the main image, or if the sky color differs, or if objects are in different positions — the image was edited after the thumbnail was generated.
This method is binary: either the thumbnail matches or it doesn't. When it doesn't match, you have direct visual evidence of what was changed. The limitation is that some editors do regenerate thumbnails, and web-optimized images often have no thumbnail at all.
Putting It All Together
No single method gives you certainty. The workflow that works best is to start broad and narrow down.
Begin with the Authenticity Checker — it runs automated metadata inspection, compression analysis, and software detection in one pass. If it flags potential issues, dig deeper with ELA and JPEG ghost analysis. If those highlight a specific region, run clone detection on that area. Cross-check the noise pattern for consistency.
The more methods that independently flag the same region, the stronger your case. A single ELA highlight could be a false positive. An ELA highlight plus a clone detection match plus a noise inconsistency in the same area? That's a pattern.
For photos you receive in messaging apps or social media, keep in mind that platforms re-compress and strip metadata before delivery. You're working with a degraded copy, which limits what forensic tools can detect. When possible, request the original file — it carries far more forensic information than a re-compressed version. Our guide on detecting edited photos covers additional techniques beyond what's listed here, and the image forensics guide provides the full technical background.
Common Questions
Can I tell if a photo was Photoshopped just by looking at it? Sometimes — warped lines, mismatched shadows, unnatural skin smoothing, and blurred edges near pasted objects are common giveaways. But skilled edits are invisible to the naked eye, which is why forensic tools exist.
Does Photoshop leave traces in the file? Yes. The Software metadata tag records the application name and version. XMP history may log every save. But metadata can be stripped, so absence of these tags doesn't mean the photo is unedited.
What about AI-generated images? Different problem. Photoshop edits start from a real photo; AI images are synthetic from scratch. The AI Detector uses different signals — texture patterns, frequency anomalies, neural network classifiers — to identify AI-generated content.
Which method is most reliable? None by itself. The strongest evidence comes from stacking multiple independent methods. When ELA, clone detection, and noise analysis all flag the same area, you can be confident something was altered.
Trust Your Tools, Not Just Your Eyes
The human eye is good at noticing when something "feels off" but bad at explaining why. Forensic analysis bridges that gap. Start with the Authenticity Checker for a quick automated scan, then use specialized tools to investigate anything it flags. Seven layers of analysis, each catching what the others miss — that's how you move from suspicion to evidence.
