Ideogram 4.5 Launches to Tackle Image Edit Drift

Ideogram has launched its Ideogram 4.5 model across its web application, API, and partner platforms, with plans to release open weights in the near future. On the Artificial Analysis AA-Image-Editing v2.0 leaderboard, the model debuted at number 23, placing behind HiDream-O1-Edit-1.5 and HunyuanImage 3.0 Instruct, but ahead of ByteDance’s Seedream 5.0 Lite. For text-to-image generation, the model ranks at number 34, marking an improvement from Ideogram 4.0 Quality, which previously held the number 40 spot.

The update specifically targets edit drift, which is the accumulation of artifacts, color shifts, and pixel degradation that typically occurs during successive image revisions. To combat this, Ideogram 4.5 supports multi-turn editing that preserves unchanged pixels. The model allows developers to use up to four reference images, input prompts up to 10,000 characters, apply optional masks, and generate native 2K outputs with dimensions reaching up to 3K. It also introduces a crop-and-stitch feature, allowing users to edit a specific high-resolution crop and merge it back into the original image without downscaling the entire file.

Pricing for the new model features a 27.5-fold spread between its tiers. The high-quality tier costs $0.22 per edit and $0.10 per generated image, while the lowest tier starts at $0.008, allowing developers to balance budget and output quality. For practitioners, these capabilities are highly suited for workflows requiring controlled, iterative revisions on approved assets. This includes changing text inside advertisements without regenerating the background, swapping out architectural finishes, or localizing product campaigns. By offering a dedicated editing stage, the model helps teams avoid the cumulative drift often seen when using alternative tools like GPT Image or Nano Banana.