Flux Kontext guide
Edit Images With Instructions, Not Layers
Flux Kontext is an image editing model that works from a picture and a written instruction. Instead of rebuilding the scene, you describe one change and review the result while the rest of the image stays familiar.
Upload
Choose a JPG, PNG, or WebP image up to 10MB.
Describe
Ask for one clear visual change in everyday language.
Review
Compare the result with the source and refine it.
What Flux Kontext changes about image editing
A conventional editor gives you a large set of controls and leaves the sequence of steps to you. Flux Kontext starts with the outcome. You point at an existing image, explain the result you want, and let the model produce a new version. This is useful when the job is easy to describe but tedious to perform: changing a background, adjusting the colour of a garment, removing a distracting object, or trying a new visual mood.
The important part is context. The model receives both the pixels in your source image and your instruction. A prompt such as “change the jacket to dark green” gives it a target without asking you to redraw the person or describe every other detail in the frame. You can inspect the result and send another instruction if the first version is close but not quite right.
This workflow is not a replacement for every part of a professional editor. It is a fast way to get from a source image to a direction worth reviewing. Precise typography, pixel-level retouching, and repeatable production templates still belong in a tool built for those controls.
Why conversation is useful for photo changes
Describe the outcome
You do not need to remember which menu hides the control. Say what should look different.
Start from your image
Your own photo supplies the subject, framing, and context instead of a blank canvas.
Review before use
Compare each result with the source. AI output can be inaccurate and still needs a human check.
Practical edits that fit a conversation
The best first request is usually a visual change you can inspect at a glance. Product teams use instruction-led editing to test a cleaner product backdrop or a new lighting direction. Creators use it to move an idea closer to a mood board without rebuilding the whole composition. Portrait retouching can start with a specific request such as removing a small object or changing the colour of clothing, then continue with a second instruction if the result needs refinement.
Background and setting
Ask for a plain studio wall, a different location, or a cleaner backdrop while keeping the subject in place.
Colour and material
Change a jacket, a product finish, or the overall colour balance when the request names the area clearly.
Objects and distractions
Remove a small unwanted element or add a simple prop. Complex structural changes still need close review.
Style direction
Explore a visual treatment such as a cartoon-like finish without making a new image from a blank canvas.
Keep the purpose of the image in view. A product photo may need a faithful label and consistent lighting, while a social post can tolerate a more expressive result. The model does not know which details are contractually important, so inspect logos, text, hands, faces, and any object that must remain unchanged before publishing the output.
How to write a better editing instruction
A useful instruction has a subject, a change, and enough context to make the target unambiguous. “Make the jacket dark green” is more actionable than “improve the photo.” The first names what should change and what the result should look like. The second leaves both the area and the desired outcome unclear.
Start with one change
Reviewing one edit at a time makes it easier to see what worked. A second instruction can handle the next adjustment.
Name the area precisely
Say whether you mean the background, the shirt, the left side of the frame, or a specific object. Avoid vague references when several elements could match.
Describe the outcome
Mention the colour, mood, material, or level of realism you want. That gives you a clearer standard when you compare the result with the source.
Refine after you see it
If the image is close, describe the remaining difference. A focused follow-up is often easier to judge than a completely new prompt.
Short instructions are not a guarantee of a perfect result, and a longer prompt is not automatically better. Write for clarity, inspect the returned image, and keep the version that serves the task. If a request is rejected by the safety checks, start with a different input or instruction rather than repeating the same request.
Flux Kontext, manual editors, and text-to-image tools
These tools solve different jobs. A conversational editor is strongest when you have a source image and a change that can be stated clearly. A manual editor is stronger when every pixel has to be controlled. A text-to-image model is the right starting point when no source image exists.
| Task | Instruction-led editing | Manual editor | Text-to-image |
|---|---|---|---|
| Keep the subject and change one detail | Describe the change in a sentence | Select or mask the area first | Rebuild the scene from a new prompt |
| Explore several directions quickly | Send a follow-up instruction | Repeat the edit across layers | Write several new prompts |
| Control exact pixels and typography | Useful for broad visual changes | Best fit for precise production work | Limited by the generated result |
How to edit an image on KontextChat
- 1
Upload the source image
Open the editor and select a JPG, PNG, or WebP file. The maximum upload size is 10MB. The preview shows which image the session will use; choose a new file if it is not the right one.
- 2
Write one clear instruction
Name the element and the change. Short, specific directions are easier to review than a long list of unrelated edits. If the result is close, send a follow-up instead of starting over.
- 3
Compare and download
Review the returned image against the source. Keep the version that fits your use, or refine it with another instruction. The download link gives you a file you can take into your own workflow.
What model does KontextChat use?
The current image editing model is @cf/black-forest-labs/flux-2-klein-4b, accessed through Cloudflare Workers AI. The fallback is a Stable Diffusion image-to-image model when the primary model is unavailable. Image generation uses @cf/black-forest-labs/flux-1-schnell. Model availability can change as we maintain the service.
Frequently asked questions
What is Flux Kontext?
Flux Kontext is a family of image editing models from Black Forest Labs. It works from an existing image and a written instruction, so you can ask for a change while keeping the rest of the scene as close to the source as possible. KontextChat provides a browser interface around the current editing model: upload a photo, describe the change, review the result, and download what you need.
Is KontextChat the official Flux Kontext product?
No. KontextChat is an independent third-party tool. The underlying models are provided by Black Forest Labs, but KontextChat is not affiliated with, endorsed by, or sponsored by Black Forest Labs.
Can I use images edited with Flux Kontext commercially?
You may use lawful results for personal or commercial purposes, subject to your rights in the input image, third-party rights, applicable law, and our Acceptable Use Policy. The Terms of Service describe the current usage terms; review them before publishing a result or using it in a product.
Do I need ComfyUI or a desktop editor?
No. KontextChat runs in the browser. Upload an image, write an instruction, and review the returned image without installing ComfyUI or a local model workflow.
Can I try Flux Kontext editing for free?
Yes. The service includes a free daily pool, subject to the limits shown in the interface and availability. When the free pool is used up, one-time credit packs are listed on the pricing page.
Powered by FLUX.1 Kontext. KontextChat is an independent third-party tool and is not affiliated with, endorsed by, or sponsored by Black Forest Labs.
Try a change on your own image
Start with the free daily pool and see whether an instruction-led edit fits your workflow. You can compare the result before spending credits.
Open the Flux Kontext editor