No More ‘Hallucinations’ or Color Shifts: Google Alumni Unveil Ideogram 4.5 for Pixel-Precise Retouching

Ideogram, a startup founded by former Google Brain researchers and the creators of Imagen, has released its generative model Ideogram 4.5. The main problem with today’s AI editors is image “drift”: ask one to change the color of a jacket or the text on a mug, and the algorithm also alters the person’s facial features, blurs textures and repaints the background. The new version tackles this pain point for commercial designers: the system can now make pinpoint edits without physically touching the rest of the scene.

The key addition is Precise Edit mode. Unlike standard inpainting, where a portion of the frame is regenerated along with adjacent areas, Ideogram 4.5 isolates changes at the pixel level. Everything outside the mask is simply copied from the source file, allowing dozens of edits in a row without artifact buildup or loss of the original resolution.

Working with 24-megapixel files and references

The model’s feature set is aimed at studio production, according to Ideogram’s official website. The algorithm can:

• recolor objects with accurate recalculation of shadows, highlights and color reflections on nearby surfaces;
• swap materials (for example, turn matte fabric into gloss or leather);
• edit or translate text inside an image while fully preserving the original font, kerning and perspective;
• use up to four reference images at once for precise style control.

High resolution got special attention. Users can crop and zoom into a small detail of a heavy photo, edit it locally and seamlessly blend it back into the overall shot. As a demonstration, the developers show processing of 4016 × 6016 pixel images (about 24 MP).

Introducing Ideogram 4.5, the most precise edit model. With each edit, leading models add artifacts, pixel shifts, and color changes. Ideogram 4.5 eliminates artifact buildup, making multi-turn editing possible. Live in Ideogram, the API, and launch partners. Open weights soon. https://t.co/cQZYfPBAw6

— Ideogram (@ideogram_ai) September 30, 2026

API pricing and how it compares with rivals

The creators position Ideogram 4.5 against OpenAI’s GPT Image 2.5 and Google’s Nano Banana lineup, claiming superior color stability across long edit sequences. In independent testing by Quantslant, the model did show minimal degradation of the source image after a chain of 10 consecutive modifications to a single frame.

The tool is already available in the platform’s web interface and via API. According to the developer documentation, processing an image at Full HD resolution costs between $0.022 and $0.205, depending on rendering depth and the selected mode.

For the digital graphics industry, the release matters because it marks a paradigm shift. AI models are gradually ceasing to be generators of “random pretty pictures” from scratch and becoming precise working tools that let designers make client revisions without risking damage to an approved layout.