Ideogram says its new model can edit part of an image without messing up the rest

Ideogram has launched Ideogram 4.5, its latest image generation and editing model, with a focus on a persistent challenge in AI image editing - making changes to one part of an image without inadvertently altering the rest. Maintaining consistency in unedited regions has long been a weak point for diffusion-based models, which tend to reinterpret the entire image during generation rather than isolating only the targeted area.

The model ships with native 2K resolution output, which puts it above many competitors that require separate upscaling steps to reach comparable quality. Pricing starts at 0.8 cents per image, positioning it as a practical option for developers and teams building image-heavy products rather than just individual creators.

On the integration side, Runway, Pika, and Leonardo AI are listed as early partners, suggesting the model is being designed with third-party workflows in mind from the start. These platforms span video generation and creative tooling, so Ideogram 4.5 could end up embedded in a fairly wide range of production pipelines relatively quickly.

Ideogram has also confirmed that an open-weight release is planned, which would allow developers to run and fine-tune the model outside of Ideogram's own platform. That detail is likely to matter to teams with specific deployment or data-privacy requirements, and it broadens the potential use cases well beyond what a closed API alone would support.