4 November 2026 · 7 min read
Film-Grain AI Photos: How to Tell a Fake That Looks Like a Snap
AI images now copy grainy, casual phone and film photos. The visual tells that survive the grain, how to check provenance, and where every method falls short.
You tell them apart by checking what the grain cannot hide: hands, any text in the frame, reflections and shadows, repeated textures, and where the image came from. AI images styled as casual phone snaps or old film photos — a bit blurry, flash-lit, grainy, badly framed — are built to slip past the 'too glossy' instinct people developed for earlier AI art. The imperfection is a costume. Look past it, check the file for provenance markers, run a reverse image search, and accept that sometimes you will not be able to know for sure.
Why do AI photos suddenly look like bad phone pictures?
Early AI images had a recognisable look: poreless skin, perfect lighting, saturated colour, everything slightly too clean. Newer image models can be prompted for the opposite — 'disposable camera', '2000s digital camera with flash', 'candid phone photo', '35mm film, grain, light leaks'. We read imperfection as authenticity, because real life is badly lit and nobody frames a party photo properly. Grain, blur and heavy compression also smudge the fine detail where many artefacts used to show.
What visual tells survive the film-grain look?
| Check | What to look for |
|---|---|
| Hands and fingers | Count the fingers, follow the joints, check rings and nails. Much better than it used to be, still worth a look. |
| Text and signs | Shop signs, labels, T-shirt slogans, number plates, posters. Garbled letters, words that look right but mean nothing, lettering that changes style mid-word. |
| Reflections | Mirrors, windows, sunglasses, puddles. Does the reflection show what should be there, from the right angle? |
| Shadows and light | Shadows pointing different ways, a flash-lit face with no flash shadow behind it, a background lit from somewhere else. |
| Repeated textures | Brickwork, leaves, crowds, hair. Patterns that repeat too neatly or melt into each other. |
| Background people | Faces that dissolve into shapes, merged limbs, two people with the same face. |
| Physical logic | Mismatched earrings, glasses arms that vanish, straps that go nowhere, chair legs that do not add up. |
| The grain itself | Real grain sits over everything, including sharp and blurry areas. Added grain can look oddly uniform. Subtle, and unreliable on its own. |
Zoom in on the largest version you can find, not a thumbnail. One odd detail is a hint — real photos have weird moments too. Several together are a pattern.
How do you check an image's metadata and Content Credentials?
Some AI tools embed C2PA Content Credentials or write generator tags into the file's metadata. When those are present, they are strong evidence of where the image came from. Skrubly's AI Image Checker reads them in your browser and tells you what is embedded; the file is not uploaded.
- Markers found: the file itself says which tool created or edited it. That is about as close to a straight answer as you will get.
- No markers found: this tells you nothing either way. Screenshots, re-saves and most social platforms remove metadata. A clean file is not proof of a real photo.
- Camera data present (make, model, settings): weak evidence on its own. EXIF can be edited by anyone with the right software.
Check a suspicious image for Content Credentials and AI generator tags — free, in your browser.
Open the AI Image CheckerHow do you trace where a photo came from?
- Run a reverse image search with Google Lens, Bing Visual Search or TinEye. Try a crop of the main subject as well as the full image.
- Find the earliest appearance. Who posted it first, and does that account usually post AI art or design work?
- Look for other angles of the same moment. Real events tend to be photographed by more than one person; a viral 'candid' shot with no second angle anywhere deserves suspicion.
- Check the account that shared it: brand new, posting in bulk, only viral imagery, no ordinary personal posts.
- Read the replies and quote posts. Someone has often already found the source or the original prompt.
Can an AI image detector tell you for sure?
No tool can tell you with certainty. Pixel-based AI detectors give a probability, not a verdict; they produce false positives and false negatives, and grain, compression and resizing — exactly what the film look adds — make their job harder. Treat any score as one input among several.
Skrubly's checker is deliberately the other kind of tool: it does not guess from pixels, it reports markers that are actually embedded in the file. That makes it reliable when markers are present and silent when they are not. Neither approach replaces looking carefully and checking the source.
A 60-second routine before you share
- Ask whether it matters. If you are about to share it, get angry about it, buy something because of it or believe a news story because of it — check.
- Zoom in: hands, text, reflections, the background.
- Save the image and check it for Content Credentials.
- Reverse search it and look for the earliest source.
- Look for a second photo or a second credible source for the same event.
- Still unsure? Say so when you share it, or do not share it. 'I can't verify this' is a perfectly good answer.
The full guide to checking any image, glossy or grainy.
How to tell if an image is AIFrequently asked questions
- Is every vintage-looking photo fake? No. Film cameras, disposable cameras and old digital cameras are popular again, and phone filters add grain too. The look alone proves nothing.
- Can I remove the grain to reveal the AI underneath? Not reliably. Denoising adds its own artefacts and can make a real photo look synthetic.
- If the checker finds no marker, is the photo real? Not necessarily. Missing markers are common on real and AI images alike, because metadata is so easily lost.
- What about AI-edited real photos? Many genuine photos now pass through AI tools for denoising or object removal. A marker can mean 'edited with AI', not 'invented by AI' — read what it actually says.