Table of Contents
A dramatic flood photo. A politician “caught” saying something shocking. A celebrity video that looks a little too perfect. Before you forward it on WhatsApp or repost it, pause. AI tools can now create or alter images and videos that fool most people at first glance.
No single check can settle this. Some AI images have obvious flaws. Others are flawless. Real photos can also look “wrong” because of compression, editing or bad lighting. This guide gives you seven checks to combine, and tells you plainly what each one cannot prove.
1. Look for visual inconsistencies
Start by zooming in. Common clues include:
- Hands and fingers: extra or merged fingers, odd joints
- Faces and eyes: mismatched reflections in the eyes, unnatural teeth or skin
- Text: garbled or meaningless words on signs, labels and clothing
- Lighting and shadows: shadows pointing in different directions
- Backgrounds: warped buildings, repeating faces in crowds
- Reflections: mirrors and windows that don’t match the scene
The catch: These clues are not proof. Motion blur, lens distortion and heavy compression can create the same effects in real photos, and newer AI images often have none of these flaws. Judge the whole scene, not one odd detail.
2. Watch videos frame by frame
Replay the clip slowly, and look at the audio too. Warning signs include:
- Faces that subtly change shape between frames
- Jewellery, glasses or clothing that appear, vanish or change
- Objects that merge or pass through each other
- Lip movements that don’t match the speech
- Lighting or shadows that shift for no reason
A real video can also carry a cloned or replaced voice, so listen as well as watch. Treat these signs as a reason to dig deeper, not as a verdict.
3. Use verification tools
Google’s SynthID Detector. Google’s tool is at synthid.com. It checks whether an image, video or audio file carries Google’s invisible SynthID watermark. Google says the watermark is built into the content itself and is designed to survive resizing, recolouring and compression. For video and audio, it can point to the segments where a watermark was found. Besides Google’s own tools, it covers partner systems including OpenAI, NVIDIA and Kakao.
Checking inside the Gemini app. Signed-in users can also upload a file to Gemini and ask, “Was this created or edited by Google AI?” Per Google’s help page:
- Upload one file at a time, up to 100 MB.
- For screenshots, crop tightly around the image.
- A detected watermark means all or part of the file was created or edited with Google’s AI.
The catch: If no watermark is found, that does not mean the file is real. It may come from an AI system that doesn’t use SynthID, or the file may have been altered too much to read.
Content Credentials. Content Credentials (built on the C2PA standard) can show who created a file and how it was edited, including whether AI tools were used. Many files carry no credentials at all, and their absence proves nothing.
4. Run a reverse image search
Upload the image to Google Lens or Google Images to see where it appeared before. This is the best way to catch an old photo passed off as a new event. For videos, pause on a clear frame, screenshot it and search that frame.
The catch: An older copy proves the caption is misleading. It does not prove the image is AI-made.
5. Check the metadata
Files may store the camera model, date, editing software and sometimes location. Look at the original file, not a copy downloaded from social media.
The catch: WhatsApp, Instagram and X usually strip metadata, and screenshots carry none. Metadata can also be edited. Missing data proves nothing, and neither does the name of an editing app.
6. Trace the original source
Ask: who posted this first?
- Find the earliest version you can.
- Look for coverage from credible news outlets.
- Compare the date, weather, landmarks and location with the claim.
- Search fact-checking organisations. In India, Alt News, BOOM, Vishvas News and PIB Fact Check regularly examine viral claims, and AFP Fact Check covers global ones.
Be cautious if a post has no source, uses an alarming caption, or pushes you to “share before it’s deleted.” Many accounts posting the same clip is not independent confirmation, since they may all be copying one unverified source.
7. Treat AI-detector results with caution
Free “AI or Not” style tools can be wrong in both directions. They may flag a real photo or miss a fake. Results change with the generator used, the compression and the editing. Use them as one clue among several.
Quick comparison
| Check | What it may reveal | Limitation |
|---|---|---|
| Visual inspection | Odd details | Real media has flaws too |
| Frame-by-frame review | Changes across frames | Good AI video can look clean |
| SynthID | Watermark from supported AI systems | No watermark ≠ real |
| Content Credentials | Origin and edit history | Most files don’t have them |
| Reverse image search | Earlier appearances | Doesn’t show how a file was made |
| Metadata | Technical details | Easily stripped or edited |
| Source tracing | Whether the claim holds up | Even trusted sources err |
A simple 60-second routine
- Zoom in and look for obvious flaws.
- Reverse-search the image or a video frame.
- Check who posted it first.
- Run it through SynthID if you can.
- If you still can’t confirm it, don’t share it as fact.
FAQs
Can I tell if an image is AI-generated just by looking?
Not reliably. Visual clues help, but modern AI images can look realistic.
What’s the easiest first check?
Reverse image search plus finding the original source. Add a SynthID check where possible.
Can Gemini detect AI-generated video?
It can check for SynthID watermarks from Google’s AI tools and indicate which segments contain them. It cannot detect content made by systems that don’t use SynthID.
Do AI detectors always give the right answer?
No. Treat results as clues, not proof.
Can a real photo be mistaken for AI?
Yes. Compression, motion blur and editing can mimic AI errors.
Should I share it if I’m unsure?
No. Say it is unverified, or don’t share it.
The takeaway
Strange hands and garbled text are clues, not proof. The strongest approach combines visual checks, reverse search, source tracing and verification tools. If the evidence stays inconclusive, “could not be verified” is a better label than “fake.”
Sources: Google (The Keyword announcement on the SynthID Detector, 7 October 2026); Google Gemini Help, “Verify Google AI-generated images and videos with SynthID”; Google DeepMind, SynthID documentation; Coalition for Content Provenance and Authenticity (C2PA) / Content Credentials; Social Media Today and Android Authority coverage of Gemini’s video verification.
Corrections: if you spot an error or an outdated detail, contact us and we will update this article.