agi.jetzt
DEEN

Model landscape

Image & Design

Models whose main task is to generate, edit and lay out images. The class is moving from "make me a picture" to controlled design work.

Last checked: 10 Oct 2026

What this class is

An image model turns a description, a sketch or an existing image into a new image. What separates the current models from earlier ones is control: editing a single region, keeping details stable across several edits, placing elements by coordinates, setting type correctly, splitting an image into layers.

There are two ways to build such a model. One is a model of its own, trained for images. The other is a general multimodal model that outputs images as one of its abilities. Both appear in this selection, and the page says which is which.

What is it optimized for?
For what an image looks like and for control over it: composition, style, legible type, and edits that change one thing and leave the rest alone.
How does it differ from a generalist?
A generalist answers in text and may call an image model as a tool. An image model’s result is the image itself. Some of the newest image models are built on general multimodal models — the line is becoming blurred.
What does it output?
Images: new ones from a description, or edited versions of existing ones. Some also return layers, transparent backgrounds or vector graphics.
Where is it useful?
Design drafts, product and advertising images, illustrations, editing existing photos, layouts with text.

Five notable models

A selection with reasons. No ranking, and no quality scores for images: agi.jetzt does not grade pictures.

  • OpenAI

    GPT-Image-2.5

    Availabledate not stated by the source
    Purpose
    Generates images from text and edits existing images.
    What is notable
    OpenAI splits its image models into a variant for precise editing (Sunburst) and a fast variant (Flare).
    Why it is here
    An image model line of its own beside the language models, with separate variants for precise editing and for speed.
    Access
    Through the OpenAI API, in two variants: Sunburst and Flare.
    Input → Output
    Text and images → Image

    Sources: OpenAI API docs: GPT-Image-2.5 Sunburst, OpenAI API docs: Image generation guide · checked 10 Oct 2026

    Model page in the registry: GPT-Image-2.5 Sunburst →
  • Google

    Nano Banana 2.1

    Availablereleased 6 Oct 2026
    Purpose
    Generates images and edits them step by step in conversation.
    What is notable
    Not a standalone image model: according to its model card it is based on Gemini 3.6 Flash and outputs images and text.
    Why it is here
    Stands for a different design: image generation as a capability of a general multimodal model instead of a separate image model.
    Access
    Through the Gemini API (model ID gemini-nano-banana-2.1).
    Input → Output
    Text and images → Image and text

    Sources: Gemini API changelog, Google DeepMind: Nano Banana 2.1 model card · checked 10 Oct 2026

    Model page in the registry: Nano Banana 2.1 →
  • Black Forest Labs

    FLUX 3 Image

    AvailablePreviewdate not stated by the source
    Purpose
    Generates and edits images, with layout control through bounding boxes.
    What is notable
    Part of a model family that, according to the provider, learns jointly from images, video and audio.
    Why it is here
    A different architecture: image is one output of a multimodal model that also generates video.
    Access
    Through the Black Forest Labs API; for companies also under a commercial weights licence. The documentation calls FLUX 3 a preview model.
    Input → Output
    Text, reference images, bounding boxes → Image, natively in 2K and 4K according to the provider
    • LimitationThe documentation calls FLUX 3 a preview model.

    Sources: Black Forest Labs: FLUX 3 Image, Black Forest Labs: FLUX 3 announcement (23 July 2026), Black Forest Labs docs: FLUX 3 · checked 10 Oct 2026

    Model page in the registry: FLUX 3 Image →
  • Ideogram

    Ideogram 4.5

    Availabledate not stated by the source
    Purpose
    Design work with type and layout; 4.5 edits images over several steps.
    What is notable
    Ideogram presents 4.5 as an editing model meant to preserve details across several edits. 4.0 is driven by structured JSON prompts.
    Why it is here
    Design as the primary purpose rather than photorealism, and a design model with open weights.
    Access
    In Ideogram, on every plan according to the documentation. Its predecessor Ideogram 4.0 has also been published with open weights since 3 June 2026.
    Input → Output
    Text and reference images → Image

    Sources: Ideogram: Ideogram 4.5, Ideogram docs: Models, Ideogram: Ideogram 4.0 technical details (3 June 2026) · checked 10 Oct 2026

    Model page in the registry: Ideogram 4.5 →
  • ByteDance

    Seedream 5.0 Pro

    Availabledate not stated by the source
    Purpose
    Controllable image generation and editing for design work.
    What is notable
    Splits an image into separately editable layers and accepts sketches as an editing reference.
    Why it is here
    The behaviour of a design tool inside the model itself: layers instead of one flat image.
    Access
    Through BytePlus, ByteDance's cloud offering for companies.
    Input → Output
    Text, images, sketches → Image, also as separate layers
    • Vendor claimAccording to the provider, stable text in the image in 14 languages.

    Sources: BytePlus: Seedream 5.0 Pro · checked 10 Oct 2026

    Model page in the registry: Seedream 5.0 Pro →

No rankingThe order is the order of this list. agi.jetzt does not grade these models and does not compare them with one another.

Why is this relevant to AGI?

To build an image as instructed, a model has to represent objects, their position in space, type and how things relate. That touches abilities which matter for general intelligence too: spatial understanding, knowledge about the world, following precise instructions.

Image models with layers and layout also show how AI moves from single answers to work products that someone can keep editing.

What this does not prove about AGI

A convincing picture is not evidence of understanding: how good an image looks says little about whether a model can reason.

Relevance to AGI does not mean demonstrated progress toward AGI. These models are not part of the frontier models, and nothing on this page enters AGI Progress.

Sources

All sources read on 10 Oct 2026.