Skip to content

Image

Flux Kontext Pro

Edit and restyle images with Flux Kontext Pro.

Available on Ofrai — chat, images, and video in one place with shared credits.

19 creditsModel details

Image studio

Preview

No output yet

Enter a prompt and hit Generate. Results will appear here.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the Flux Kontext Pro creator above.

How it works

Create with Flux Kontext Pro in three steps

  1. 01

    Choose your inputs

    Keep Flux Kontext Pro selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Image options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

Flux Kontext Pro is the engine for work that starts with a picture you already have. It reads the supplied image, keeps everything you did not ask it to change, and applies the change you described. The interesting part is not that it can edit — several engines can — but that editing is a different job from generating, and this is the one built around the edit.

For an asset that exists and needs adjusting, a restyle, a replacement, or a correction, editing is cheaper, faster and more predictable than regenerating, because the parts of the picture you have already approved are the parts that get carried forward untouched.

The difference between editing and generating

Generation starts from nothing and invents a whole frame, which means every decision is open and every decision can come out differently. Editing starts from a specific frame and changes a specific thing, which means the result is bounded by what you supplied. That bound is the value.

A product shot that is right except for the background, a scene that needs the logo swapped, a set of frames that needs a grade applied consistently, a style that needs changing without changing the subject — in all of those the framing, the subject and the composition are already decided and should not move. Regenerating them throws away the work you have already done, and the new version has to earn your approval again from scratch. Editing keeps the approval and changes the part you named.

What actually gets preserved

The value of an edit is entirely in what it leaves alone, so the rule to learn is that you do not have to describe the rest of the picture. Name the change, and everything else survives: the subject, the pose, the composition, the lighting, the framing. If you find yourself re-describing the subject to be safe, you are fighting the engine rather than using it, and the result is usually a new image that happens to resemble the old one.

The discipline that makes editing productive is the same one that makes generation productive: one change per pass, and look at what moved. Because each edit starts from the current state, a sequence of single-change edits converges, while a sequence of multi-change edits just produces a different result each time and tells you nothing about which instruction caused what.

Restyling, and describing a look properly

Restyle is the capability that gets bought for this engine and the one that is easiest to get wrong. The instruction has to describe the target look as a set of properties, not as a mood. Naming a decade and a medium, the character of the light, the material of surfaces, the rendering style and the colour relationships gives the model things to move toward, and it will land somewhere recognisable. A mood word gives it nothing to hold.

The practical test is whether your instruction could be checked: if you cannot say how you would know the target look was hit, the instruction is too vague to be useful, and the render will come back as a plausible frame in some other style.

Edits that work, and edits that fight the picture

The constraint to plan around is that some changes can be made in place and some cannot. Replacing an object, changing a colour, restyling, removing or swapping a background, applying a consistent grade — these are all in-place changes, and this engine does them well. Changing the camera angle, moving the subject, altering the light direction while keeping the rest, turning a portrait into a landscape — these are not.

When a request fights the supplied image, the picture usually wins, and you get a partial edit: the change is applied somewhere that does not quite work, and no error is reported. The rule is simple enough to state and worth stating: edit what can be edited in place, and regenerate when the change is structural.

Where the edit loop belongs in the workflow

The productive sequence puts this engine at the end of a pipeline rather than in the middle of it. Generate the base frame on whichever engine suits the look you are after, then bring it here to restyle, to place the real product, to correct the one thing that came out wrong, or to run a set of variations off a frame you have already approved.

The reason to sequence it this way is that an edit is cheaper and far more predictable than trying to get a generating engine to land a specific detail on the first attempt, and because the edit inherits a composition that was already signed off rather than proposing a new one. For a set, the practical benefit is bigger still: you approve one frame, then change one variable per variation, and the set stays coherent because the variations are edits of the same approved frame.

Prompting an edit

State the target, name the constraint, and stop. First, what should change — a specific object, a specific area, a specific property. Second, what must not change, if anything is at risk, since a reference edit can still drift if the change is ambiguous. Third, and this is the one people skip, what the result is for, in one clause: whether the frame is going into a banner, a product page, a lookbook.

That clause is what lets the model choose a treatment that fits the destination instead of one that merely looks good in isolation. Keep it to a few sentences. A long edit prompt invites the model to satisfy the prose and quietly lose the frame, and then you are back to generating.

FAQ

What does Flux Kontext Pro keep from the image I upload?
Everything you did not name as a target — the subject, the pose, the composition, the lighting and the framing all survive. If you find yourself re-describing the subject to be safe, the instruction is fighting the engine instead of using it, and the result is usually a new image that resembles the old one.
Can I use it to restyle an existing photo?
Yes, and that is what it is built for. Describe the target as specific properties — the period and medium of the light, the rendering style, the material of surfaces — rather than as a mood. If you cannot say how you would recognise the target look, the instruction is too vague to land.
When should I regenerate instead of editing?
When the change is structural: moving the subject, changing the camera angle, or altering the light direction while keeping everything else. This engine was not built to re-stage a scene, and asking it to produces something plausible and wrong. Edit what can be edited in place.

FAQ

Can I use Flux Kontext Pro directly on this page?
Yes. The creator above is the working Ofrai studio with Flux Kontext Pro selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with Flux Kontext Pro?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Image models that can handle similar creative work to Flux Kontext Pro.

Commercial use depends on the selected model and model terms.