Image inpainting editor in n8n with FLUX.1 Fill

Image inpainting editor in n8n with FLUX.1 Fill

An n8n workflow that serves a small image editor in the browser and sends each edit to Black Forest Labs' FLUX.1 Fill.
You open the webhook's address, pick or load an image, paint over the part to change, describe what should appear there
and press Generate. The result opens beside the original in a before-and-after view.

What is inside

• workflow.json: the workflow as n8n exports it, nineteen nodes. Seven are sticky notes, and one of them was added by
  Mindket. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• FLUX Fill called api.bfl.ml, an older address of the API that Black Forest Labs' docs no longer use. It now posts
  to https://api.bfl.ai/v1/flux-pro-1.0-fill, as the docs show (read on 12 November 2026).
• Check FLUX status asked get_result?id= on that old address. It now polls the polling_url that the first call
  returns, which the docs ask for on the main endpoint.
• safety_tolerance was 6, the least strict level on a scale where 0 is the strictest. It is now 2, the API's default.
  Raise it in FLUX Fill if moderation blocks your own edits.
• The editor's script and style loaded from a public repository on jsDelivr by branch name, so anything pushed there
  later would run in your page. They now load from one fixed commit. The two libraries from unpkg are pinned to exact
  versions, Konva 9.3.22 and img-comparison-slider 8.0.7.
• Mockups listed five example photos kept in the original author's file storage. It holds one placeholder,
  <YOUR_IMAGE_URL>.
• The links to the author's credential on the two BFL nodes and the ids of two webhooks are gone.
• A sticky note says what the two BFL nodes now call.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Black Forest Labs API key.
• Optional: a few public image addresses for the editor's list. The editor can also load a file from your computer.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Header Auth credential whose name is x-key and whose value is your BFL key. Select it in FLUX Fill and in
   Check FLUX status.
3. Open Mockups. Replace <YOUR_IMAGE_URL> with the address of an image your browser may load from another site, or
   delete the entry.
4. Save, open Webhook and listen for a test event, then open its test address in a browser. To use it for real,
   publish the workflow and open <YOUR_N8N_HOST>/webhook/flux-fill.

How a run goes

1. Opening the address sends a GET. Webhook hands it to Mockups, which builds the list of images, and to Merge,
   which joins that list with the webhook's own data, including the page's address.
2. Editor page fills the list and the address into the HTML, and Respond to Webhook returns it as a page.
3. Generate sends a POST to the same address, with the image, the mask, the prompt, the steps, the guidance and the
   prompt upsampling choice. The POST output of Webhook goes through NO OP to FLUX Fill.
4. FLUX Fill cuts the data-URL header off the image and the mask and posts them to BFL as a PNG request.
5. Wait 3 sec, Check FLUX status and Is Ready? loop every three seconds until the status is Ready.
6. Get Fill Image downloads the result from result.sample, and Show the image to user returns it to the page with
   its own content type. The page shows it against the original, and the buttons reuse or save it.

Change the result

• The images in the editor's list: Mockups.
• The model settings: the Steps and Guidance sliders on the page, and output_format and safety_tolerance in
  FLUX Fill.
• How often it polls: Wait 3 sec.
• Saving the result somewhere: add nodes after Get Fill Image, as the sticky note there suggests.
• The page itself: the HTML in Editor page.

Good to know

• BFL's docs, read on 12 November 2026, say a black mask area is kept and a white one is inpainted. The editor draws the
  mask for you.
• Since n8n 1.103.0, HTML that a webhook returns renders in a sandboxed iframe. n8n's docs say scripts there cannot
  reach the parent window, relative addresses fail and the headers for Basic Auth are missing. The editor uses an
  absolute address, and the webhook's Allowed Origins option is * by default.
• Because of that last point, Basic Auth on Webhook would not reach the page's own POST. The webhook has none, as in
  the original copy, so anyone who learns its address can spend your BFL credits. Unpublish the workflow when you are
  not using it, or set the node's IP allowlist.
• Any status other than Ready, including a failed or moderated request, sends the loop round again until the workflow's
  120-second timeout ends the run.
• BFL's docs say result addresses expire after 10 minutes, do not send CORS headers and are not for serving to end
  users. That is why the workflow downloads the image itself.
• Each Generate is one paid BFL request.
• The page needs to reach unpkg and jsDelivr to load its script and style.
• Picking the placeholder image in the list shows an error until you replace its address.
• Your image, mask and prompt go to Black Forest Labs.