GPT Image 2 Artifact Cleaner

Removes the artifacts GPT Image 2 leaves behind — excessive sharpening, random bright specks, unnatural edges, scale-like texture — while leaving the image itself intact. Trained specifically for GPT Image 2; it will not transfer usefully to other generators.

Drop an image, then drag the divider to compare. Differences are easiest to judge at full resolution — use the fullscreen button on the result.

Input

Cleanup strength (α)

Raise it if artifacts remain; lower it if the image goes soft or texture smears. Among the settings that look right, prefer the lowest.

0.25 · conservative 0.5 · balanced 0.75 · strong 1.0 · aggressive

Before / after

Examples

Every image passes through the FLUX.2 VAE, so untouched areas are not a pixel-exact copy of the input. Some detail loss is inherent: the artifacts sit at the same scale as real texture. Judge by how the image looks, not by how much was removed.

ZeroGPU gives visitors a few minutes of GPU time per day — for batches, run the CLI locally. Noncommercial use only (PolyForm NC 1.0.0).