Concern about deepfakes is justified, but it distracts from the fact that the overwhelming majority of misleading images involve no advanced technology at all.
The Categories, From Common to Rare
- Miscontextualized real photos. A genuine, unedited photograph presented as showing a different event, place or date. This is by far the most common form of image misinformation, and it requires no editing skill whatsoever.
- Misleading crops. Removing the person or object that changes the meaning entirely — the classic example being a crop that hides who someone is actually talking to.
- Selective editing. Brightness, colour, removal of a small object, or choosing the one frame in a burst where someone looks angry.
- Composites. Combining elements from different images.
- Synthetic generation. A wholly generated image of an event that never occurred.
- Deepfakes proper. Face or voice replacement in video, or synthetic video of a real person.
Note the shape of this list: effort increases going down, frequency decreases. Your verification habits should be weighted accordingly.
Context Beats Forensics
For a typical person evaluating a typical image, these questions catch far more than pixel analysis:
- Where did it first appear, and when? A reverse image search revealing the same photo from three years ago settles most cases immediately.
- Who is sharing it, and what's their record?
- Does anyone else document the same event? Real events produce multiple independent images from different angles.
- Does the caption make a claim the image alone can't support? Very often the photo is real and the *sentence* is the falsehood.
- Does it feel designed to make you angry? Emotionally activating content spreads faster and is more likely to be manipulated or miscontextualized.
Technical Checks Worth Doing
- Reverse image search — the single highest-value tool.
- Check Content Credentials if present.
- Look for compression inconsistency: a region that's noticeably sharper or softer than the rest, mismatched noise, or edges that are too clean.
- Check shadows and reflections for physical consistency.
- Beware detector tools' confidence. They give probabilities, produce false positives on legitimate edits, and fail on screenshots.
If You Are Targeted
Non-consensual manipulated imagery of real people is a serious harm, and increasingly illegal:
- Document everything — URLs, screenshots, dates, account names — before reporting, since content often disappears.
- Report to the platform under its synthetic or non-consensual imagery policy; most major platforms now have specific categories for this.
- Seek legal advice. Many jurisdictions have introduced specific offences for non-consensual synthetic imagery, alongside existing defamation and harassment law.
- Contact support organizations that specialize in this; you don't have to manage the process alone.
Protecting Your Own Images
You can't prevent misuse entirely, but you can raise the cost:
- Limit high-resolution public photos of your face where you can.
- Watermark images you publish professionally.
- Keep originals with intact metadata — they're evidence of what the authentic version looks like.
- Enable Content Credentials where your tools support it.
- Review privacy settings on accounts where photos of you accumulate.
The realistic posture is neither panic nor complacency: assume images can be fabricated, verify anything consequential through sources rather than inspection, and treat provenance as the thing worth checking.