The strangest photo trend of 2026 is that people want their pictures to look worse. Not badly composed, not out of focus. Worse in a very specific way: grainy, slightly faded, a little imperfect. After three years of AI generators flooding every feed with flawless, plastic-smooth images, imperfection has become the new way of saying a human was here.
I spent an afternoon building and testing a film grain filter for ImgCruncher, and this post is the result: what the anti-AI look actually is, the exact two-filter recipe I use to get it in the browser, and the file size surprise that nobody warns you about. Every number in here is a real measurement from a real test file, and every screenshot is the actual tool.
Why everyone suddenly wants imperfect photos
Scroll any feed and you can spot the AI images without trying. They are too clean. Perfectly smooth skin, perfectly graded skies, not a speck of noise anywhere. Generators produce mathematically clean pixels because there was never a physical sensor or a roll of film in the loop, and our eyes have quietly learned to flag that cleanliness as synthetic.
So creators started running the process in reverse. Film grain, faded blacks, muted color, light leaks, the whole analog toolbox is back, and this time it is being applied on purpose to digital and AI images alike. Trend roundups this summer keep calling it the anti-AI aesthetic or intentional imperfection, and it is showing up everywhere from album art to product photos. The logic is simple: grain reads as texture, texture reads as physical, physical reads as real.
The irony is worth naming: people generate an image with AI, then spend another five minutes making it look like AI had nothing to do with it. I am not here to judge. I am here to show you the fastest way to do it.
What makes a photo read as AI in the first place
Before adding anything, it helps to know what you are covering up. Three things make an image feel generated:
- Zero noise. Real cameras always produce some sensor noise, especially in shadows. Generated images have none, so large flat areas like skies and walls look airbrushed.
- Perfect gradients. A sunset from a generator transitions with mathematical smoothness. A photographed sunset has subtle banding, noise and color wobble.
- Cranked color. Models are trained on images people liked, and people like saturated images, so generators lean vivid by default. The result sits somewhere between HDR photography and a postcard.
My test photo for this article is honestly a perfect specimen: an over-processed HDR mountain lake that looks the way AI images look, even though it is a real photograph. Teal water, glowing orange peaks, not a grain of noise in sight. If I can make this look analog, anything you throw at the filter will be easier. If you want the full checklist of AI tells, I wrote a separate guide on how to tell if an image is AI-generated.
The recipe: fade first, then grain
The anti-AI look has two halves. Faded color handles the "too vivid" problem, and grain handles the "too clean" problem. Doing only one of them gets you halfway; the combination is what sells it.
Open the photo filter tool and drop your image in. You get all 13 looks rendered as live thumbnails of your own photo, so you pick by eye:
Here is the exact sequence I used on the test photo:
The tool applies one filter at a time by design, so layering means download and reload. That sounds like a limitation until you realize it is also the control you want: you judge the fade on its own before grain goes over the top of it.
Pass one: Fade at 70. Fade lifts the blacks and mutes the saturation, which kills the postcard color instantly. At 100 percent it can look washed out, so I pulled the intensity slider back to 70. The photo went from screensaver to something a film stock might have produced.
Pass two: Grain at 60. The Grain filter adds luminance noise, the same noise in all three color channels, which is how film grain behaves. It also fades the image slightly on its own, so it stacks naturally on top of the Fade pass. At 60 percent the grain is clearly visible without turning the photo into static:
The before and after, side by side:
How strong should the grain be
The intensity slider blends the effect with your original from 0 to 100, and the right value depends on where the image is going. Here is the same photo region at 200 percent zoom so you can see what the grain actually does to pixels:
My working ranges after testing a pile of values:
- 35 to 50: subtle. Viewers will not consciously notice grain, but the image stops feeling sterile. Good for product shots and thumbnails.
- 55 to 70: the visible film look. This is the trend range, what you see on the faux-analog posts all over Instagram right now.
- 80 to 100: heavy, deliberately rough. Works for moody black and white, especially stacked on the Noir filter instead of Fade.
One practical note: grain shows more on dark images than bright ones, because noise is more visible in shadows. If your photo is mostly dark, start 15 points lower than you think you need.
Grain makes your file bigger. A lot bigger
Here is the part no trend post mentions. JPEG compression works by predicting pixels from their neighbors, and random noise is by definition unpredictable. Add grain and you have made your image dramatically harder to compress. The numbers from my test chain:
| Stage | File size | What happened |
|---|---|---|
| Original photo, 1100×741 | 161.2 KB | Starting point |
| After Fade 70 | 186.3 KB | Small bump from the re-encode |
| After Grain 60 | 350.4 KB | More than double the original |
That is a 117 percent size increase for the exact same resolution. On a full-resolution phone photo the same effect can push a 2 MB image past 5 MB, which matters when platforms cap uploads or recompress aggressively.
The fix is one pass through the image compressor with a target size. I gave it the grainy 350.4 KB file and a 200 KB target:
The result kept the grain fully visible at 195.0 KB, a 44 percent reduction. Quality 82 is high enough that the grain structure survives; the compressor finds that value automatically by binary-searching the quality range against your target.
Compress as the last step, always. If you compress first and add grain after, you get both compression artifacts and doubled file size, the worst of both worlds. Filter first, grain second, compress last. If the photo is headed for a specific platform, the Instagram sizes guide has the exact dimension and size targets to aim at.
What this will not do
Honesty section, because the analog aesthetic is a deep rabbit hole and a browser filter covers a specific slice of it:
- No light leaks or halation. Those orange streaks and glowing highlights you see in some film emulations need positional effects, not per-pixel ones. The filter tool does not fake them.
- No scratches, dust or borders. Same reason. If you want a full fake-film-scan look with sprocket holes, you need a dedicated film emulation app.
- Uniform grain only. Real film grain clumps differently in highlights and shadows. This filter applies even luminance noise, which at normal web viewing sizes reads the same, but a film purist pixel-peeping at 400 percent could tell the difference.
- One filter per pass. Layering means download and reload, as covered above. It is three extra clicks, not a dealbreaker, but you should know before you start.
Everything runs in your browser, nothing gets uploaded, and it works on a phone. For a trend look that most people will view for 1.5 seconds on a small screen, that trade is easy. And since your photos never leave the device, this is also the safe way to process personal pictures; I wrote about why that distinction matters separately.
The 60-second version
- Open the photo filter tool and drop in your photo, AI-generated or real.
- Click Fade, set intensity around 70, download.
- Load the download back in, click Grain, set 55 to 70, download again.
- Run the result through the compressor with a target under your platform's limit.
- Post it and let people wonder what camera you used.
If you came here from the AI side and want the reverse workflow, prompts that produce good raw material in the first place, my AI image prompt guide covers that end to end, and the viral AI photo trends post tracks what people are actually posting right now.
