Ideogram 4.5 Joins Text-to-Image, and Reference Images Get Smarter

Hi there,

Two updates for AI image generation this week: a new model built for precise edits, and reference images that now make the most of each model.

New model: Ideogram 4.5

Ideogram 4.5 is now available in the Text-to-Image model list.

Ideogram's models are known for their crisp typography, and this new version focuses on precise editing: change one part of an image, like a color, a piece of text or an element of the scene, while the rest stays untouched. Details also hold up when you edit the same image several times in a row, so you can refine a visual step by step without it drifting away from the original.

That makes it a great fit for product photos to retouch, signs or packaging text to adapt, and any visual where a small change shouldn't ruin everything else. You can add up to 5 reference images to guide it.

Select it from the model picker in the AI toolbar. As always, its credit cost is shown before you generate.

Reference images: the right number for each model

Until now, every Text-to-Image model was limited to 4 reference images, whatever it could actually handle. Many models can take a lot more, and some take fewer.

Abyssale now applies each model's real limit:

• Up to 16 images with GPT Image 2.5 Sunburst, GPT Image 2.5 Flare and GPT Image 2
• Up to 14 with Nano Banana 2, Nano Banana Pro and the Gemini models
• Up to 10 with GPT Image 1.5, 8 with FLUX.2 Pro, and 7 with Seedream
• Fewer for a few lighter models, like Nano Banana (3)

In practice, that means you can give the AI much more to work from: several product angles, your brand's visual style, a logo and a mood board, all in a single generation. The result is closer to what you have in mind, in fewer attempts.

The same limits apply when you generate through the API.