Flux 3 Image Brings Multi-Step Edits That Keep Scenes Intact

Black Forest Labs (BFL) has released Flux 3 Image, the image-generation part of its new Flux 3 model family. The main selling point is editing. According to BFL, users can make several edits in a row without the model changing parts of the picture they never asked it to touch.

That is a narrow claim, but an important one. Most people who refine an AI image over several rounds want each change to stay contained, so that fixing a background does not also alter a face or a product label. BFL says Flux 3 Image handles that kind of step-by-step workflow. So far, that comes from the company itself and has not been independently tested.

What the model covers

Flux 3 Image is a general-purpose model, not a dedicated editing tool. BFL lists four core capabilities:

1. Text-to-image: generating pictures from written prompts.
2. Image-to-image: transforming or editing an existing picture.
3. Text rendering: placing readable text inside images.
4. Photorealism: producing realistic photographic output.

There are also controls for people who need more than a single prompt. Users can lay out a scene with bounding boxes, which means drawing boxes to mark where specific elements should appear. They can add up to ten reference images to steer the result, and output goes up to 4K resolution.

A free demo is available for anyone who wants to try the model before paying for it.

Pricing, licensing and open weights

BFL is offering several ways to access the model, depending on how much control a customer wants.

• API access is discounted by 50 percent through October 8.
• Commercial weights can be licensed by companies that want to run the model and fine-tune it on their own infrastructure.
• An open-weight version is expected "in the coming weeks", according to the company. BFL has not given a firm date.

The commercial weights license matters for companies with strict data requirements. Running a model in-house means images and reference material do not have to pass through an outside API. It also makes fine-tuning on internal assets possible.

Ideogram moves in the same direction

BFL is not alone. Shortly before Flux 3 Image launched, Ideogram announced version 4.5 of its own model, which also focuses on editing. Ideogram says this model will be released as open weights soon as well.

Two image-model developers chose the same focus within a short time. That suggests the competition is moving from "who generates the prettiest single image" toward "whose model is most useful across a full editing session."

Our Take

The Flux 3 Image launch says more about where image AI is heading than about any single feature. Generating a striking picture from one prompt is no longer enough to stand out. Controllable, repeatable editing is what professional workflows need, and that is what BFL is selling. If its claim holds up, Flux 3 Image could be more useful for design and marketing work than for one-off experiments.

The open-weight plans from both BFL and Ideogram also stand out. This mirrors a broader trend toward running capable models locally rather than relying only on cloud APIs.

There are open questions. Higher photorealism makes convincing fakes easier, a concern that recent research on AI avatars fooling people has already raised in a different medium. It is worth watching whether independent testing confirms BFL's editing claims. Two other questions are when the open weights actually arrive and under what license terms.