Ideogram has released version 4.5 of its image model in its web app, API and partner platforms. The model focuses on precise, repeated image edits while preserving composition, colors and untouched pixels. It also offers tools to change stylized lettering or translate text while retaining the surrounding layout and visual styling.
Successive edits with fewer unwanted changes
The central problem the new model targets is image degradation across successive edits. Artifacts can accumulate, colors can change, details can lose sharpness and pixels can shift as revisions continue. Ideogram says version 4.5 preserves the original image more reliably through these sequences than GPT Image and Nano Banana.
The text tools can change stylized lettering directly where it appears in an image. They can also translate text while preserving the surrounding layout and visual treatment.
API workflows for precise edits and reference-based generation
The API gives developers two main workflows. Precise edit returns an edited image with the same dimensions as the input. Generate and edit instead combines generation guided by reference images with editing instructions.
A request can include up to four reference images and an optional mask. A high-precision mode is also available, designed to preserve pixels outside the area being edited.
For high-resolution images, developers can process only a crop and merge the edited result back into the original, avoiding the need to downscale the entire image for a localized change. The model also offers native 2K output, with supported dimensions reaching approximately 3K.
Ideogram 4.5 offers four quality modes and accepts prompts of up to 10,000 characters.
Artificial Analysis ranks it 23rd in editing, with strengths in text and composition
Ideogram 4.5 placed 23rd on Artificial Analysis’s AA-Image-Editing v2.0 leaderboard. It ranked behind HiDream-O1-Edit-1.5 and HunyuanImage 3.0 Instruct, but ahead of ByteDance’s Seedream 5.0 Lite. It is the first Ideogram model listed on that leaderboard, and Artificial Analysis evaluated it in High quality mode.
In Artificial Analysis’s evaluation, the model comes closest to the leaders in three areas. Text and symbol editing includes changes to lettering within images. Scene and style editing covers adjustments to lighting or backgrounds, while composition and framing includes cropping and changing the apparent camera viewpoint.
Generation improves over Ideogram 4.0 Quality in Artificial Analysis results
For text-to-image generation, Ideogram 4.5 ranked 34th in Artificial Analysis’s evaluation, compared with 40th for Ideogram 4.0 Quality. The new version narrowed the gap to the leaders in five of nine evaluated capabilities, with its largest gains in text rendering and complex compositions.
The model’s strongest categories in this evaluation are Knowledge, Physics and Complex Compositions. Knowledge covers recognizable landmarks and species, while Physics tests physical relationships, including gravity and support. Complex Compositions assesses whether images follow the counts, positions and assigned attributes specified in a prompt.
Among the evaluated use cases, version 4.5 showed its largest improvement over version 4.0 in Animation and Gaming. These generation results, like the editing evaluation, correspond to High quality mode.
Available in the app and API, with model weights planned for release
The model can be accessed through Ideogram’s web app, its API or partner platforms. In the API, High quality mode costs $0.22 per edited image and $0.10 per text-to-image generation; the lowest tier costs $0.008 per edit and $0.03 per generation.
Ideogram also plans to release the model’s weights.
