OpenAI has announced ChatGPT Images 2.5, an updated model for creating and editing images. The key difference from version 2.0 is improved editing: now the model can precisely modify individual elements of a scene without redrawing everything around them, and it handles long chains of edits within a single conversation much better. After several iterations, the image stays more consistent, and previous changes remain stable.

In addition, generation is now about 50% faster. Visual quality has also improved: the neural network renders textures and lighting more accurately, and when using reference photos, it better preserves the appearance and distinctive features of objects.

Images 2.5 also gained support for transparent backgrounds and became more reliable at handling complex compositional requests that involve multiple objects, their relative positions, and numerous conditions.