Ideogram Releases 4.5 Image Editor, Claiming Less Drift Across Repeated Changes

Ideogram has released Ideogram 4.5 on its platform and through its developer API, promising that changing one part of an image will leave the rest alone. The image-editing model targets repeated revisions: Ideogram says it reduces the unwanted shifts in pixels, colors and textures that can accumulate as users keep adjusting the same picture.

The release is built around preserving an image, rather than simply producing a new version of it. On its model page, Ideogram describes that accumulated change as drift. Its stated goal is to let users refine an image over several rounds without losing details they wanted to keep.

That promise connects several of Ideogram’s highlighted uses: changing product imagery, adjusting an interior, restoring an old photograph or replacing text inside a design. These are editing tasks where the requested change may be narrow, even when the source image contains many other details.

The competitive claim is about the revision sequence

Ideogram’s comparison page applies the same sequence of edits to its model, GPT Image 2.5 Sunburst, Nano Banana Pro and Nano Banana 2. The company claims the competing outputs become unusable within a few edits, while Ideogram 4.5 maintains cleaner results. That is Ideogram’s assessment of its own side-by-side examples, not an independent verdict.

The distinction is important to the launch’s pitch. The comparison asks whether an image survives a chain of revisions, rather than whether one isolated edit looks good. Ideogram’s stated advantage is continuity across those requests: fewer color changes, displaced pixels and unwanted texture artifacts as the edits continue.

A crop-based route to high-resolution edits

Ideogram also says 4.5 can edit high-resolution images without downsizing the original. Its described workflow isolates a region, edits that crop and preserves the crop’s edges so it can be stitched back into the source image. The promise is both local precision and a clean boundary between the changed area and its surroundings.

The company illustrates this with a 4,016-by-6,016-pixel source image, labeled 24.2 megapixels. It positions the technique for close-ups and large-format printing, where preserving sharp details and avoiding a visible edit boundary are central to the result. The example demonstrates the workflow Ideogram is promoting; it is not a guarantee for every source image.

The edits Ideogram highlights

• Color and lighting: refine the appearance of an existing image, including new product colorways.
• Text modification: change words inside an image as a targeted editing task.
• Restoration and transformation: restore old photos, work from sketches or use a style reference.

Hosted access now; two distribution steps remain planned

The Decoder reports four quality tiers, all at native 2K resolution. That output specification is distinct from Ideogram’s high-resolution source-editing example: the latter describes editing a crop and reinserting it into a larger original, not generating an entire 24.2-megapixel image in one pass.

Recraft separately confirmed that Ideogram 4.5 was released this week and said it is coming to Recraft Studio. Its October 1 announcement describes a workflow in which users generate with their preferred model, then bring in Ideogram 4.5 for careful revisions. Recraft has not given an arrival date.

An open-weight release—making the model’s trained weights available—is also planned, according to The Decoder’s account of Ideogram’s announcement. The timing is described only as coming soon. For now, the confirmed access routes are Ideogram’s own platform and API; downloadable weights and Recraft integration remain future steps.