GPT Image: image generation and editing with OpenAI
GPT Image is OpenAI's family of models for generating and editing images from text and images in context, through an API. What the iterative process brings, references as input, and the place of such an engine in a pipeline, and in the direction Nova is taking.
GPT Image refers to OpenAI's family of models for image generation and editing: they take text and images in context to create or modify a visual, through an API. Their value is not only an image from a sentence, but the iterative process (generate, observe, ask for a change, keep, correct), close to art direction. At Axiris, GPT Image is a possible step in a pipeline, picked up afterwards in the finishing tools; Nova is designed to connect to it when a project calls for it. That is the direction announced in September 2026, not yet an open service.
Image generation and editing with OpenAI
GPT Image refers to OpenAI's family of models for image generation and editing. These models take text and images as context to create or modify a visual. They are used through an API: no hardware is needed on your side, and the images you send are processed on OpenAI's infrastructure, under its terms.
From prompt to visual conversation
The value of a multimodal model is not limited to producing an image from a sentence. It allows an iterative process:
generate → observe → ask for a change → keep → correct
This logic is close to what you look for in an art direction workflow. It has a limit, shared by every engine that reinterprets the whole image at each request: what was good can change along with what you are correcting. That is why Axiris works by areas and freezes what is validated. See Building the image by areas.
References as input
Images serve as input to give the model context: a photograph, a concept, a composition, graphic elements, existing references. How you give each reference a role so that the engine does not mix them up: Workflow with visual references.
GPT Image in a pipeline
Depending on the project, GPT Image is one step among several. A produced image is then picked up in Photoshop, in ComfyUI or in other creation and finishing tools: selection, retouching by area, upscaling.
The engine does not replace the workflow. The other engines that can edit, and what sets them apart: Image editing models.
GPT Image and Nova
Nova, the Axiris AI, was designed around a local infrastructure; she is not defined by the model running behind her. The direction announced on 5 September 2026 is to make Nova an orchestration layer that calls on different engines according to the task: a local model when it is the most relevant, a specialised service when it brings a particular capability, and the large models through APIs, including OpenAI's, when they bring real value to the workflow: NOVA becomes the heart of AXIRIS.
Nova is therefore designed to connect to GPT Image when a project calls for it. At the time of writing, this is not an open service; we do not promise it before it is. How the choice of engine will be made, and under which confidentiality conditions: Axiris is preparing its AI router (6 October 2026) and Local AI model or API?.
Official sources
Image Generation guide: OpenAI · Model documentation
Contact request: a first conversation, free and with no commitment. Every practice is different; it is by listening to you that we propose what suits you.