Ideogram 4.5 focuses on multi-pass editing without image degradation

Ideogram's new model limits pixel, color, and texture drift from one edit to another, and retouches high-resolution images without shrinking them.

Ideogram has released version 4.5 of its image model, focusing on editing rather than generation. The problem is well known to users: each new instruction tends to move pixels slightly, alter shades or introduce artifacts, so that after a few round trips the image becomes untapped.

Ideogram asserts that its version 4.5 significantly reduces this drift and publishes comparisons with GPT Image 2.5 Sunburst, Nano Banana Pro and Nano Banana 2 on the same series of modifications, where its competitors degrade after a few laps. The company highlights uses such as color variations, restoration of old photos or targeted retouches that leave the rest of the image intact.

Thought for large format printing

Another novelty: the model works on high resolution images without subsampling them. It preserves the edges of a cropped and modified area so that it can be reinserted seamlessly into the original, which aims at close-ups and large format printing.