The AI look is a real thing and everyone can see it

You know it when you see it. Skin that's suspiciously smooth. Lighting that's too even, too golden, too perfect. Eyes with that slightly glassy, "I've seen the face of God and it was a loss function" quality. Backgrounds packed with detail that no real photographer would include. And the overall vibe of something that was designed by committee to be generically beautiful.

That's the AI look, and it's becoming as recognisable as a stock photo watermark. People scroll past it. Clients clock it. Your design friends definitely notice it and are absolutely judging you.

The good news: it's mostly fixable. AI image generators default to producing AI-looking images, but with the right prompts and some post-processing, you can make output that genuinely passes for human-created work. Here's how.

The tells (know your enemy)

Before we fix the problem, let's name it. These are the things that scream "AI made this":

Too-smooth skin. Every AI model defaults to porcelain skin with no pores, no blemishes, no texture. It looks like someone applied three layers of foundation and then Gaussian-blurred the result.

The "AI shine." A weird, uniform glossiness on surfaces that shouldn't be glossy. Fabric that looks slightly wet. Wood that looks lacquered when it should be matte. Metal that has too-perfect reflections.

Impossibly perfect lighting. Real photos have imperfect lighting. Shadows go in slightly wrong directions. Light sources are inconsistent. AI images often have this omnidirectional, soft, golden-hour glow that looks beautiful and completely fake.

Overly detailed backgrounds. In real photography, backgrounds are either out of focus or intentionally minimal. AI images tend to cram detail into every pixel. A simple portrait ends up with a background that looks like a detailed matte painting.

Symmetry. Real faces aren't perfectly symmetrical. Real buildings aren't perfectly aligned. Real compositions aren't perfectly balanced. AI defaults to symmetry because mathematically it's "optimal," and that's exactly why it looks wrong.

The hands thing. This has improved dramatically but AI still occasionally produces hands with extra fingers, missing joints, or impossible poses. Always check the hands.

Fix 1: Prompt for imperfection

The single most effective technique is explicitly requesting imperfection in your prompts.

Instead of: "A portrait of a woman in a garden, beautiful lighting, high quality"

Try: "A candid photo of a woman sitting in a garden, natural light, overcast day, slightly out of focus background, shot on Fujifilm X-T4, 56mm lens, f/2.8, minor lens flare, visible skin texture"

The camera and lens references are powerful because they force the AI to mimic the specific characteristics of real camera systems. A "shot on Fujifilm X-T4" image has different colour science than a "shot on Canon R5" image, and both look more real than a generic AI render.

Key phrases that help across all models: - "natural light" or "available light" (kills the golden AI glow) - "candid" or "unposed" (reduces the portrait studio look) - "slightly overexposed" or "slightly underexposed" (real photos aren't perfectly exposed) - "visible grain" or "film grain" (adds organic texture) - "imperfect composition" (reduces the centred, symmetrical default)

Fix 2: Use negative prompts aggressively

If you're using stable-diffusion, flux, or leonardo-ai, negative prompts are your best friend.

Negative prompt template: "smooth skin, airbrushed, perfect lighting, HDR, oversaturated, studio lighting, centred composition, symmetrical face, bokeh balls, lens flare, perfect teeth, Instagram filter, stock photo"

Yes, you're telling the AI to avoid things that are traditionally considered "good" in photography. That's the point. Those things, in combination, are what make images look AI-generated. Real photography is full of technical "imperfections" that give images character.

Fix 3: Reference mundane lighting

This is the trick that makes the biggest visual difference with the least effort. Stop asking for dramatic lighting.

midjourney in particular defaults to dramatic, cinematic lighting if you don't specify otherwise. It loves golden hour, volumetric light rays, and rim lighting. These look gorgeous but they immediately register as "too perfect" to the human eye.

Instead, reference boring lighting: - "Overcast afternoon, flat lighting" - "Fluorescent office lighting, slight green cast" - "Harsh midday sun, strong shadows" - "Kitchen lighting, slightly warm, one overhead source"

Real photos are often shot in boring lighting because real life happens in boring lighting. By matching this, your AI images immediately feel more grounded.

Fix 4: Specify a photography style, not an art style

Vague: "a photo of a coffee shop, beautiful, detailed"

Specific: "a photo of a coffee shop interior, documentary photography style, shot on Leica Q2, 28mm, natural window light, customers slightly motion-blurred, 1/60 shutter speed, ISO 800 grain visible"

Naming a specific camera, focal length, and technical settings gives the AI a concrete visual reference. The result looks like it was taken by a photographer, not generated by a computer.

Prompt templates that work:

For portraits: "[subject description], editorial portrait, shot on Sony A7III with 85mm f/1.4, shallow depth of field, natural window light from the left, visible skin texture and pores, slightly asymmetrical pose, muted colour palette"

For food: "[dish description], overhead shot, natural daylight from window, slight shadow on one side, on a slightly worn wooden table, crumbs visible, napkin partially in frame, shot on Fujifilm X-T4"

For architecture: "[building description], street photography, shot on Ricoh GR III, 28mm, harsh daylight, pedestrians slightly blurred, signage visible, power lines in frame"

For products: "[product description], product photography, single softbox from upper left, grey seamless background, slight shadow beneath, shot on Canon R5 with 100mm macro, f/8"

Fix 5: Post-processing in Photopea or Lightroom

Even with perfect prompts, AI images benefit from post-processing that makes them feel less "digital."

Open the image in Photopea (free) or Lightroom and: Add grain. Filter > Noise > Add Noise. Amount 2-4%, Gaussian, Monochromatic. This single step makes images feel dramatically more like photographs. Adjust the curves slightly. Lift the blacks a tiny bit (drag the bottom-left point of the curves up). This mimics the look of slightly faded or unprocessed photos. The crushed blacks of AI images are a giveaway. Crop imperfectly. AI images are always perfectly framed. Crop so the subject is slightly off-centre, or cut off a bit of the edge of something. Real photographers do this constantly because real compositions aren't perfect. Reduce saturation by 10-15%. AI images are often slightly oversaturated. Pulling the saturation down makes colours look more like real-world colours and less like HDR processing. Add a subtle colour cast. Real photos have colour casts from their lighting. Add a very slight warm cast (for indoor shots) or cool cast (for overcast outdoor shots) using the colour balance or temperature slider.

Fix 6: Avoid the default Midjourney look

midjourney has a recognisable aesthetic. Everything looks like concept art for a film that cost $200 million but went straight to streaming. Moody, atmospheric, dramatic, beautiful, and obviously AI.

To break out of the Midjourney look: - Add "--style raw" to your prompts (reduces Midjourney's aesthetic processing) - Specify "--no dramatic lighting, cinematic, atmospheric, moody" in your negatives - Use "--stylize" at lower values (try --stylize 50-150 instead of the default) - Reference specific photographers rather than styles. "In the style of Martin Parr" or "in the style of William Eggleston" gives much more human-looking results than "cinematic" or "editorial."

The uncomfortable truth

Even with all of these techniques, someone who works with images professionally can probably still tell your AI images aren't photographs. The technology is getting better incredibly fast, but we're not yet at the point where AI images are truly indistinguishable from real photos at full resolution.

What these techniques do is raise the bar significantly. Instead of "obviously AI" at first glance, you get "might be AI if I really look closely." For social media, blog posts, and most commercial uses, that's more than good enough.

For anything where authenticity genuinely matters (journalism, documentary, legal), please don't use AI images and pretend they're real. That's a different conversation entirely, and not one that ends well for anyone.