GPT Image 2.5

OpenAI's image model, in two variants, for creating and editing: multiple references, partial or full editing, transparent background. What it does, and how it arrives in the Axiris interface.

GPT Image 2.5 is the current generation of OpenAI's image models, available through the API in two variants: one for editing precision, one for fast everyday generation. According to OpenAI's documentation, the model creates an image from a text, modifies an existing image in part or in full, combines several reference images and renders a transparent background. In Axiris, it arrives as a tool of the interface: you say what you want, Nova helps you put it into words, the model generates, and the result comes back into the studio to be picked up again area by area.

OpenAI's image model

GPT Image 2.5 is the current generation of OpenAI's image models. Its documentation names two variants: Sunburst, presented as the most capable model for generation and editing, to be preferred when editing precision matters most; and Flare, for fast, good-quality everyday generation. The family, and OpenAI's place in a creative pipeline, are presented in GPT Image.

What it does

According to OpenAI's documentation:

Create an image from a description.
Edit an existing image, in part or in full, and chain modifications over several turns.
Edit an area by providing a mask. OpenAI specifies that this mask guides the model without guaranteeing that it follows the exact shape.
Combine references: one or more images serve as the starting point for a new image.
Render a transparent background, in PNG or WebP.
Write in the image: text rendering has clearly progressed, and OpenAI notes that it can still lack precision in placement and sharpness.

What Axiris does with it

This model's strong point for us is how it reads a layout. Give it a precise composition, a poster built by areas, a reference mock-up, and it renders an image faithful to that structure, text included. That is the move we are preparing: you build the layout in the studio, Box Director, Nova helps you write what each area expects, GPT Image renders the image, and the result comes back into the studio.

There, the model's limits are compensated for: a mask it follows only approximately becomes an area you delimit yourself, and what is validated no longer moves. Building the image by areas and Image editing models.

Nova knows what this model expects and what it does less well than others. For the same job, she can point you towards it, towards another cloud model or towards a studio engine. How to choose an AI image model?

What you need to know

GPT Image is a remote model: the images and the instructions go to OpenAI, under its terms. You decide, project by project, whether a job can leave the studio: Local AI model or API?. A generation is paid in Axiris credits. Credits

Where access stands

Access to GPT Image 2.5 from the Axiris interface is in preparation. This page describes what the model does, according to OpenAI's documentation, and what Axiris will do with it. The opening will be announced in the news.

Official sources

OpenAI, Image generation guide · OpenAI, model documentation

Contact request: a first conversation, free and with no commitment. It is by listening to you that we propose what suits you.