AI News Feed
Market watch
Products & Applications

ChatGPT Images 2.5 Draws Complaints Over Noise and Grain

OpenAI's ChatGPT Images 2.5 is drawing complaints over persistent grain and noisy artifacts from Reddit users and TechRadar tests.

OpenAI says the new image model preserves people and objects from reference photos better, generates images up to 50 percent faster and follows complex visual instructions more reliably. It also introduces Sketch for turning rough drawings into finished images, templates for posters and product photography, and the ability to leave comments directly on an image when requesting edits. OpenAI calls it its new state-of-the-art image model.

TechRadar reported that the early reaction has been far from tidy. The complaint centers on fuzz that can make images look gritty or sandy even when no camera was involved. TechRadar said noise is especially damaging because it can masquerade as detail. Generative models learn that photographs contain grain, texture, tiny color variations and imperfections, and Images 2.5 appears unusually eager to reproduce those signals across skies, skin, feathers and studio backdrops. One user wrote that OpenAI needs to get rid of these noisy artifacts because they make the model mostly useless for production. Praise was mixed in with the complaints, but it fell short of universal acclaim.

TechRadar tested several images with highly detailed subjects and areas that should remain visually smooth, using prompts that asked for smooth gradients and the elimination of visual noise. In a photo-studio mug image, the peacock, flowers, fruit and landscape wrapped around the ceramic were impressive at first, but the grey studio background had a persistent granular texture and some of the smallest patterns began to dissolve into a mess. TechRadar said the model generated a great deal of visual information without always deciding which parts deserved clarity.

A barn owl at twilight produced a similar result. The feathers were handled well, with individual structures visible across the wings and face, convincing talons and none of the plastic quality that older AI wildlife images often had. Behind it, however, a sky that should have looked like a clean sheet of paper instead showed obvious fine grain across a smooth blue area. TechRadar said the grain did not ruin the picture, but once noticed it was hard to ignore.

A fantasy scene with a winged horse or dragon, a rainbow and an alpine lake was coherent in composition, but the issue was visible immediately. TechRadar said the relentless crispness gave the image the air of a video game loading screen. The pale-blue sky had a faint textured quality rather than a completely clean gradient, and the mountains, trees and creature details had a crunchy, heavily sharpened look when examined closely. TechRadar described it as nowhere near a disaster but similar to an overprocessed photograph.

The strongest test, two friends on a rooftop in the evening, came closest to the promised leap in realism, with convincing expressions and clothing that seemed affected by real gravity. After looking for more than a minute, though, grain across the twilight sky, hair and skin became glaringly obvious. TechRadar noted that there was no poorly set lighting or malfunctioning camera sensor, so a grainy twilight sky appeared to be a creative decision or model artifact rather than the unavoidable physics of shooting in bad light.