Technique · AI content creation

Common AI image artefacts, and how to fix them.

AI image artefacts are the small physical and logical errors generators still make: extra or fused fingers, uneven teeth and eyes, garbled text, impossible reflections and shadows, objects merging into each other, warped product shapes and repeated patterns. A structured review at full resolution catches most of them, and most are fixed by regenerating the faulty area rather than the whole image.

Where artefacts hide

People: count fingers, check how hands grip objects, look at teeth, ears, eye alignment and jewellery that melts into skin. Physics: every shadow should agree with the light source, and reflections in mirrors, glass and water should match what is in front of them. Objects: chair legs, cutlery, bag straps, bicycle spokes, railings and windows are where geometry breaks. Text and logos: look for invented letters. Products: label distortion, wrong caps, changed proportions. Backgrounds: small faces in crowds are often the worst part of an otherwise good image.

A review routine that scales

Review hero images at full resolution, never as thumbnails in a grid. Check in the same order every time, for example product, people, hands, text, edges, then background, so nothing depends on where the eye happens to land. Use a short checklist for each asset type, and a second reviewer for anything that runs large or carries a claim. Reviewer fatigue is real: rotate the work, and note which kinds of artefact keep appearing so the brief can prevent them next time.

Fixing rather than starting again

Most faults sit in a small part of an otherwise approved image, so regenerate that area alone with inpainting rather than rolling the dice on a new image. If a pose keeps failing, simplify it: hands holding fewer objects, fewer people, a less complicated grip. Sometimes a crop is the cleanest fix. Finish hero assets with manual retouching where precision matters. Upscaling can introduce fresh artefacts, such as invented texture, over-sharpened edges and new detail in faces, so review again after every upscale.

Prevention at the brief

Many artefacts can be designed out before generation. Keep hands relaxed or out of frame unless they matter, avoid mirrors and glass directly behind subjects, limit background crowds, keep readable text out of the scene and add it in layout, and choose simpler props. Product distortion is better prevented than repaired: Synthetic White's product lock keeps real products pixel-accurate in geometry, labels and colours while the scene around them is generated.

Updated 26 October 2026

Questions

AI image artefacts, answered.

Why does AI still get hands wrong?

Hands take a huge range of poses and are often partly hidden in the photographs models learn from, so the model has a weaker sense of their structure than of faces. Results have improved a lot, but complex grips and hands holding several objects still fail most often.

Can artefacts cause problems beyond looking bad?

Yes. A distorted product or label can misrepresent what customers will receive, which is a misleading advertising risk, and visible mistakes in brand imagery get screenshotted and shared. Retailers and platforms may also reject low-quality or inaccurate product images.

Do small background details really need fixing?

For hero images and anything shown large, yes. Viewers zoom in on phones, and an odd face or a melted railing in the background is exactly the kind of detail that gets noticed and passed around. Small thumbnails can be held to a lighter standard.

Get started

See your AI studio generating this week.

Thirty minutes. A live studio in your colours, your brand hub loaded, a real campaign brief.

See plans