How to edit images with FLUX Kontext
Upload a source image, describe one controlled edit, and compare the result before taking another pass. Keep the instructions that survived review.
Updated 5 min read
FLUX Kontext edits an existing image from a text instruction: supply the picture, describe the change, then inspect what changed and what should have stayed the same. BFL’s Kontext editing guide (opens in a new tab) documents this in the browser Playground and through its API. Keep each accepted pass with its instruction; fenbs can hold that record so another assistant can help you continue without guessing which image came first.
One distinction matters before you begin: BFL labels FLUX.1 Kontext [pro] a previous-generation model and recommends FLUX.2 for new editing projects. This guide helps when you specifically need a Kontext workflow or are continuing one. Do not assume a tutorial mentioning Kontext describes the newest FLUX family.
What do you need before the first edit?
You need an input image and a written instruction; the editing guide requires both. BFL’s Quick Start (opens in a new tab) explains the account and credits used for browser and API work. Start with a copy of a picture you have permission to use, and keep the untouched original separately.
Choose a change that is easy to judge. For example, you might change a chair’s upholstery from beige to blue while preserving its shape and the room around it. Write the intended change and the protected features before generating, so a pleasing new picture does not distract you from a damaged detail.
- Change: the chair upholstery should become blue.
- Keep: the chair silhouette, leg positions, camera angle, wall color, and floor pattern.
- Check: the upholstery edges and shadows, plus any objects near the chair.
How do you make a controlled browser edit?
BFL’s documented browser route is to upload an image in the Playground and describe the change. For an overview of the available routes, see where to use FLUX. You do not need to build an API integration to try the editing approach.
- Open playground.bfl.ai and use the Kontext editing workflow. Confirm which model you are using before comparing it with an older test.
- Upload the source image you intend to edit. Give your local copy a clear name, such as chair-original, so you can identify it later.
- Describe one change in plain words. For example: “Change the chair upholstery to blue. Keep the chair shape, camera angle, room, and floor pattern.”
- Generate an edit and compare it against the source. Check the requested change first, then the features you wanted to preserve.
- Keep or reject that pass before asking for another change. Save the accepted file and the exact instruction beside its source filename.
These example words describe a test, not a promise that every protected detail will remain unchanged. If the room changes too much, simplify the request and return to the original source for another attempt. Do not keep building on a pass you already know is unsuitable.
How should you phrase text replacements?
The Kontext guide demonstrates text edits using quotation marks around the original wording and the replacement. An example instruction would be: “Replace ‘OPEN DAILY’ with ‘OPEN SATURDAY’ on the sign. Keep the sign’s position and the surrounding wall.” Give the requested text explicitly rather than asking for “a better headline.”
Read the output letter by letter at a useful viewing size. Inspect capitalization, spacing, repeated letters, and nearby text that should not have changed. The vendor’s examples do not guarantee perfect spelling, so a successful-looking preview still needs a person’s check before publication.
For work involving many small labels or approved brand lettering, keep the final text check as a separate task. The broader AI workflow for designers explains how to carry an accepted direction into production. Our guide for graphic designers covers evaluating tools for the whole job.
How do you compare several passes fairly?
BFL demonstrates iterative character edits, but those examples are not an identity guarantee. For a person or product, compare distinctive features with the original after every pass. A useful background change may still alter a face, logo, or product shape enough to make that version unusable.
Use a short sequence with a clear parent for each image: original to color edit, then accepted color edit to background edit. If you compare two alternatives, generate them from the same parent and name them accordingly. Otherwise, you may mistake a difference inherited from an earlier pass for the effect of the new instruction.
| Record | Example |
|---|---|
| Input and output | chair-original to chair-blue-02 |
| Instruction | Change upholstery to blue; preserve the room and chair shape. |
| Accepted change | Blue is close to the intended direction. |
| Remaining problem | Inspect the front leg edge before using the image. |
| Next action | Compare with the original at the final display size. |
Our design review guide covers checking an image against its requirements. Keep the accepted sequence and remaining checks in fenbs, so the next session knows which pass is the starting point.
What changes if you use the API?
The documented Kontext request goes to https://api.bfl.ai/v1/flux-kontext-pro with prompt and input_image. Check completion using the returned polling_url; do not invent a result address. The separate FLUX API walkthrough handles the integration steps.
After completion, download the result into durable storage. BFL’s integration guide (opens in a new tab) explains that delivery URLs expire and are not intended for direct end-user serving. Save the request recipe without its key, plus the actual input and output files, rather than treating the temporary URL as the finished handoff.
Never lose the work you did in an AI session
After a useful AI session, save it in fenbs: the key observations, findings, research and context. Then carry on from exactly where you left off, on any device and in any other AI. The next AI you connect reads it at the start of the session, so you never have to explain the same problem again.
Create a free accountFree for up to 5 people and 50 open tasks · No credit card · Works with Claude, ChatGPT, Cursor and more
How to keep accepted edits with fenbs
Keep the accepted work so you can carry on from exactly where you left off, on any device and in any other connected AI. The next assistant needs the source, the instruction, and your review, not a pile of unnamed images.
- Create a free account.
- Save the exact edit instruction, full model name, and links to the source and accepted image files.
- Write what passed, what still looks wrong, and which image the next edit should begin with.
- Connect a supported assistant from Integrations, then ask it to read the saved work before helping you write the next instruction.
- Save the new result and next step before you stop, so you can pick up the same work later.
fenbs does not edit the image or automatically collect your Playground history. An image editor and a folder may be enough for a single correction. Use fenbs when review findings, approved choices, and unfinished checks need to survive a change of assistant; find related workflows in AI at work.