Ideogram 4.5 vs Flux 3 Image: New AI Image Models

Two new image generation and editing models arrived this week: Ideogram 4.5 and Black Forest Labs' Flux 3 Image. In both you can draw boxes on the canvas to define regions. Where they differ is focus: Ideogram 4.5 leans on not degrading an image across successive edits, while Flux 3 Image leans on granular control over composition and on reference images.

This article summarizes the relevant sections of the AI Search channel's weekly AI news video on YouTube and extends them with the information on the two companies' own model pages; the video link is at the end. This is a news roundup, and I have not tested the models. The comparison examples come from the companies' own pages; they are not independent measurements.

What is Ideogram 4.5?

Ideogram presents 4.5 as "the most precise edit model." Its starting point is this problem: image models introduce small pixel shifts, color changes and texture artifacts with every edit, and after a few edits the image becomes unusable. Ideogram 4.5 aims to reduce this drift and preserve details across multi-turn edits.

Highlights from the page:

• Multi-turn edit comparison: The same edits are shown side by side with GPT Image 2.5 Sunburst, Nano Banana Pro and Nano Banana 2. According to Ideogram, rival outputs become unusable within a few edits; Ideogram 4.5 stays clean after each edit.
• Kinds of edits: color and lighting, text modification, product photography, interior design, architecture, restoring old photos, sketch to image, style reference, reframing and depth to image.
• Zoom editing: You can edit a high-resolution image without downsizing it. The edges of the edited crop are preserved, so you can stitch the piece back into the original seamlessly. The example on the page is a 4,016 × 6,016 pixel (24.2 megapixel) image.
• Bounding boxes: According to the video, as in version 4 you can draw a box on the canvas and edit only that region.

Access: For now on Ideogram's platform. According to the video, the company plans to open-source 4.5; the presenter says that if it happens it could be the best image editor you can run locally. That is an expectation; the page gives no open-weights date. Page: ideogram.ai/models/4.5.

What is Flux 3 Image?

Flux 3 Image is Black Forest Labs' model, which its page describes as "maximum control over every pixel." The same family also includes Flux 3 Video and Flux 3 Action.

• Composition with bounding boxes: You draw boxes on the canvas and write what goes in each; in the example on the page each box has a name, coordinates and a description. You can send a layout prompt in the same form to the API. You do not have to use boxes; plain text works too.
• Native 4K: The model generates directly at 4K.
• Editing: You can re-describe specific boxes and leave the rest "locked, unchanged." Several edits can be made at once; in the example a surfer's wetsuit and board turn red at the same time while the wave and sky stay as they were. Changing text inside an image without touching the unchanged text is among the examples shown.
• Reference to image: You can add up to 10 reference images; each gets a token in the order you add it, and that is how you refer to it in the prompt.
• Access: The Playground and Black Forest Labs' API. For companies running at scale there is a commercial weights license; you contact sales for it.

The video's warning still holds: weights are said to be coming for Flux 3, but the same promise was made for Flux 3 Video months ago and no open model is visible yet. Page: bfl.ai/models/flux-3-image.

Side by side

	Ideogram 4.5	Flux 3 Image
Focus	Consistency in multi-turn editing	Composition control and references
Bounding boxes	Yes (per the video)	Yes, named boxes and coordinates
Reference images	Not stated on the page	Up to 10
Resolution	Editing at any resolution	Native 4K generation
Editing	Natural language, multi-turn	Re-describe per box, others locked
Access	Platform; open source planned	Playground and API; commercial weights license

Which for what?

• If you will edit the same image many times (product photo retouching, design revisions, old photo restoration): Ideogram 4.5's consistency claim and zoom editing stand out.
• If you will build a composition from scratch (poster, book cover, multi-element scene): Flux 3 Image's named boxes and 10-reference support look like a better fit.
• Cost and access: Neither page lists prices; check the plans.

This reading rests on the companies' own descriptions. Run a few rounds of editing on your own images and compare; drift shows up exactly there.

For the agent that arrived in ComfyUI, which runs image and video models locally, see the Comfy Agent article; for installing an open-source video model, see the How to install LTX 2.5 guide.

Frequently Asked Questions

Is Ideogram 4.5 open source?

Not for now. According to the video, Ideogram plans to open-source it; no date was given.

Why is Ideogram 4.5 called "the most precise edit model"?

It reduces pixel shift, color change and texture artifacts across multi-turn edits. In Ideogram's own comparison, GPT Image and Nano Banana outputs become unusable within a few edits while Ideogram 4.5 stays clean.

How many reference images does Flux 3 Image accept?

Up to 10.

Is Flux 3 Image open source?

No, for now it is used through the API and Playground. There is a commercial weights license for large-scale use; a general open-weights promise has not been kept yet.

What are bounding boxes for?

You draw a box on the canvas and describe what goes in that region. In Ideogram only that region is edited; in Flux you build the composition region by region and can lock any boxes you like.

Source

• AI Search, Gemini 4, GPT 6.1, Dots, Claude Sonnet 5.5, Ideogram 4.5, Flux 3: AI NEWS (YouTube, October 4, 2026): the Ideogram 4.5 and Flux 3 Image sections.
• Ideogram, Ideogram 4.5: model page.
• Black Forest Labs, FLUX 3 Image: model page.