FLUX Kontext: How to Edit a Photo with AI (2026)

FLUX Kontext (full name FLUX.1 Kontext) is an image editing neural network from Black Forest Labs. Unlike ordinary "text to picture" generators, it works in "change X, keep Y" mode: you upload a ready photo and write an instruction, replace the background, change the color of clothes, remove an extra person in the back, and the model changes only what you asked for and leaves the rest untouched. In Pixyn, FLUX Kontext works right in the browser, and new accounts get welcome tokens to start.

TL;DR

• What it is: an instruction-based image editor from Black Forest Labs. It takes a picture plus a text instruction and makes local and global edits while keeping the unchanged part of the frame.
• The main difference from an ordinary neural network: it is not generation from scratch from a description but an edit of an existing image. It needs an imperative instruction ("replace", "remove", "add") rather than an artistic description of the scene.
• Versions in Pixyn: FLUX.1 Kontext Multi edits uploaded images (one or several at once); Kontext Pro (a balance of quality and speed) and Kontext Max (maximum quality, more careful with typography) create images from text.
• Strong sides: keeps the face, pose, lighting and background that you did not touch; holds the same object between edits; according to Black Forest Labs, identity "drifts" more slowly than with previous edit models over a series of consecutive edits.
• Where to use it: editing product photos for marketplaces, retouching, changing the background and clothes, updating lettering on banners.

---

What FLUX Kontext is and who is behind it

FLUX Kontext is developed by Black Forest Labs, a German team founded by the authors of the original Stable Diffusion. The same team releases the FLUX family of generators. The company announced the Kontext line on May 29, 2025 as a separate branch, built specifically for editing rather than for generation from scratch.

The key idea is in the name: Kontext (German for "context"). The model works "in context": it takes both text and the source image as input at the same time and builds the result based on what is already in the frame. According to Black Forest Labs, one model combines three things: a precise edit of a separate area, a global transformation of the whole scene and classic generation. In practice, its value is in editing: keep what is there and carefully change a part.

How an edit differs from an ordinary neural network (t2i)

An ordinary text-to-image generator (Seedream, FLUX generators) draws a picture from scratch from a text description. You write "a girl in a red dress against the background of Paris" and get a completely new image, different every time.

FLUX Kontext works differently. You give it a ready photo and an instruction about what to change in it. It does not redraw the frame anew: it takes your image as a base and edits a specific detail. "Replace the red dress with a blue one": the dress turns blue, while the face, pose, background and lighting stay the same. This is a fundamentally different tool for a different task: not "invent a picture" but "edit mine".

If you need to generate an image from a blank canvas, see the models catalog or the review of Seedream.

---

FLUX Kontext versions in Pixyn

Black Forest Labs released Kontext in several variants. The Pixyn FLUX group has three Kontext models. (There is also the open [dev], a model with open weights and 12 billion parameters under a non-commercial license, for local research; the platform does not use it, I mention it for completeness.)

FLUX.1 Kontext Multi: editing uploaded photos

This is the editor itself in the Studio. You attach one image or several and write an instruction about what to change. It suits most everyday tasks: change a color, replace a background, remove an object, adjust one detail. With several images, the model takes elements from them for one result.

FLUX Kontext Pro and Max: images from text

In Pixyn, Kontext Pro and Kontext Max create an image from a text description. Pro is optimized for the "quality and speed" compromise. According to Black Forest Labs, Max follows the prompt noticeably better and works more carefully with typography without losing speed.

The choice logic is simple: edit a ready photo with Kontext Multi; create a new picture from text with Pro, or with Max when the text in the frame matters.

---

Version in Pixyn	What it does	Strong side	Good for
FLUX.1 Kontext Multi	Edits uploaded images by instruction	Changes only what you name	Color, background, object, a detail
FLUX Kontext Pro	Image from text	A balance of quality and speed	Everyday generation
FLUX Kontext Max	Image from text	Better prompt following, more careful with text	Pictures with text in the frame

The takeaway from the table: take Kontext Multi to edit a photo, Pro or Max to create a new picture from text.

Strong sides of FLUX Kontext

Keeps what you did not touch

This is the main reason to take Kontext instead of an ordinary generator. Everything you did not name in the instruction, the model tries to leave as it was: the face, pose, lighting, background, texture. For commercial editing this is critical: reshoot a catalog in a new setting without the risk that the product's shape or the logo color "slides".

Iteration: edits in a series with slow drift

Kontext is designed for multi-round editing: you make an edit, then another one on top of the result, then another, in a series. The weak point of all edit models here is "drift": with each edit the character gradually stops looking like themselves, the face floats. According to Black Forest Labs (the study measured the cosine similarity of facial embeddings between the source and the results of consecutive edits), Kontext holds identity better than previous methods and drifts more slowly, so the margin in the number of edits is larger.

Editing text in the frame with simple instructions

Kontext can replace lettering on an image with a direct instruction like Replace 'JOY' with 'BFL', keeping the original font and position where possible. For short simple lettering, updating a word on a banner, fixing a price tag, replacing the text on packaging, it works neatly.

Speed

Black Forest Labs positions Kontext as fast compared to similar edit models: editing a ready frame goes noticeably faster than a full regeneration of the scene. For a batch of similar edits this saves time.

---

How to use FLUX Kontext in Pixyn

1. Sign up in 30 seconds via email, Telegram or Google. Create an account. New accounts get welcome tokens, so you can try it without paying.
2. Open the Studio, the "Images" section with the FLUX group.
3. Choose FLUX.1 Kontext Multi for editing.
4. Upload the source image, the one you need to edit.
5. Write the instruction: what to change and what to keep (see below for how to phrase it).
6. Click "Generate": the result appears in the feed, and you can download it or send it to a new edit.

---

How to write an instruction prompt for editing

The main rule: write an imperative, not a description of the scene. Kontext is an editor: it needs a command about "what to do", not a picture of the world.

• Bad (a t2i description): a girl in a blue dress against a Parisian street, soft light, evening. Kontext will think you want to redraw the whole frame and will break the reference.
• Good (an instruction): replace the red dress with a blue one, leave everything else unchanged.

The structure of a working instruction:

1. An action verb: replace / remove / add / change / recolor.
2. The object: what exactly (the dress, the background, the lettering, the person in the back).
3. The result: into what (blue, a studio grey background).
4. What to keep: "leave the face, pose and lighting unchanged" (especially useful so the model does not touch anything extra).

Working example 1: changing the background of a product photo

Replace the background with a clean light-grey studio backdrop.
Keep the product, its colors, shadows and reflections exactly the same.

The product stays pixel for pixel, only the background changes: exactly what you need for a uniform marketplace card.

Working example 2: editing a detail while keeping the rest

Change the model's jacket color to navy blue.
Keep her face, pose, hair and the background unchanged.

One element changes, and the model tries to leave everything else untouched. If there are several edits (color plus an accessory plus the background at once), split them into clear steps.

---

Anti-prompt: what NOT to write

These phrasings break the result precisely because Kontext is an editor, not a generator. Remove them in advance:

beautiful girl in a flowing dress, cinematic lighting, paris at night,
masterpiece, ultra-detailed, 8k
tiny dense paragraph of legal text on the label
interlaced fingers extreme close-up

Why:

• a detailed artistic description of the scene without the words "replace / remove / keep" is a t2i prompt; Kontext will try to redraw everything and lose your reference. Give a specific command.
• long dense text on a label falls apart; keep text edits short and simple.
• an emphasis on "interlaced fingers in close-up" provokes anatomical artifacts; do not focus the frame on hands.

---

Scenarios: where FLUX Kontext is really useful

Editing product photos for marketplaces

One shot frame becomes many card variants: change the background to an even studio one, recolor the product into another color of the item, remove extra objects from the table. The main object stays unchanged: you do not need a new photo shoot for each variant.

Retouching and cleaning up the frame

Remove a random person in the background, delete trash or a wire in the corner, fix a small detail: precise edits where it matters not to touch the rest of the image. Kontext keeps the surroundings untouched better than a full regeneration.

Changing clothes and background for a catalog

Recolor the model's clothes, change the background for a seasonal shoot, replace the setting, while keeping the face and pose. For a marketplace catalog this is a fast way to update the visuals without a photo studio.

Updating lettering on banners

Change a short text on a ready promo banner or packaging with a direct instruction Replace '...' with '...', keeping the font and layout. For long lettering, add the text as a separate layer.

If the task is large, dozens or hundreds of similar edits on one template, it is convenient to automate with a visual pipeline: set it up once and run it in one click. How this works is in the article What Workflow in Pixyn is.

---

What it costs in tokens

FLUX from 2 tokens per image, Nano Banana from 3. An edit is usually cheaper than generating from scratch: you do not overpay for what already worked.

The studio shows the exact charge before the run: check against it rather than the article, because the catalog gets updated. Pack contents are on the pricing page.

FAQ

How is FLUX Kontext different from an ordinary neural network for generating pictures?

An ordinary neural network draws an image from scratch from a text description. FLUX Kontext edits a ready image by instruction: you upload a photo and say what to change, and it keeps the rest. In editing mode it is an editor, not a generator.

What is the difference between Kontext Multi, Pro and Max?

In Pixyn, Kontext Multi edits uploaded images by instruction, one or several at once. Kontext Pro and Max create an image from text: Pro is a balance of quality and speed, and Max, according to Black Forest Labs, follows the prompt better and works more carefully with text.

Can I generate a picture from scratch with FLUX Kontext?

Yes: in Pixyn, Kontext Pro and Kontext Max create images from text. For other image generators, see Seedream or Nano Banana Pro.

Can I change text on an image?

Yes, for short simple lettering, with an instruction like Replace 'JOY' with 'BFL', keeping the font where possible. For guaranteed accurate long text, use Ideogram or Recraft, or lay out the text as a separate layer.

Will the face and pose be kept during an edit?

This is Kontext's strong side: everything you did not name in the instruction, it tries to leave unchanged. To be safe, add an explicit "keep face and pose unchanged" to the prompt. For big changes after a long series of edits, go back to the source image.

Can I use the results commercially?

Yes: for product photos, banners and advertising. Plans and terms are on the pricing page.

---

Ready to try FLUX Kontext?

Sign-up takes 30 seconds via email, Telegram or Google, and new accounts get welcome tokens to start.

Edit your first image in FLUX Kontext →

Already signed up? Open the Studio, choose FLUX.1 Kontext Multi, upload a photo and write your first edit instruction.

Read also: AI face swap, done through reference-based editing in FLUX Kontext and Nano Banana.

Sources

• Black Forest Labs, the developer of FLUX