Web form that draws a Flux picture in one of five styles

Web form that draws a Flux picture in one of five styles

An n8n workflow with a web form. A user writes a prompt and picks one of five styles. The workflow adds that style's
long prompt text and asks the FLUX.1 schnell model, through Hugging Face, to draw it. It saves the picture to an
S3-compatible bucket and answers with a page that shows it beside the last four renders.

What is inside

• workflow.json: the workflow as n8n exports it, 18 nodes. 13 do the work and five are sticky notes that label the
  steps. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The call went to api-inference.huggingface.co, an address that no longer resolves (checked on 12 November 2026).
  Mindket's copy posts to the Hugging Face route for fal.ai, https://router.huggingface.co/fal-ai/fal-ai/flux/schnell,
  with the prompt in a field called prompt. That route answers with a link to the picture, so a new node,
  Download image, fetches it before Upload image to S3. Its error output goes to Respond with error, like the
  other calls.
• The bucket name and the public address of the author's bucket are now <YOUR_BUCKET_NAME> and
  <YOUR_PUBLIC_BUCKET_HOST>, in Upload image to S3 and five places in Serve webpage.
• The page had a link to duplicate the author's template. It is gone. Two sticky notes, a setup video and a credit for
  the project, are gone as well, and the note about the model now says what the call does.
• The error text promised that the model fails about one time in ten, which described the old address. It now says the
  picture could not be made and to try again.
• The two credentials were linked to the author's n8n. The links are gone, so select your own. The form trigger's webhook
  id is gone too.
• The file had no settings block, so Mindket's copy adds executionOrder v1, the setting n8n gives a new workflow.
• Nothing else. The five styles, the form, the page and the S3 step are the file's own.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Hugging Face token that can call Inference Providers. Hugging Face bills those calls to your account.
• An S3-compatible bucket that serves its files at a public address, such as a public Cloudflare R2 bucket, and an S3
  credential for it.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an HTTP Header Auth credential. Give it the name Authorization and, as its value, the word Bearer, a space
   and your Hugging Face token. Select it in Call hugginface inference api.
3. Add an S3 credential with your bucket's endpoint, region, access key and secret, and select it in
   Upload image to S3. Replace <YOUR_BUCKET_NAME> in that node with your bucket's name.
4. Open Serve webpage and replace <YOUR_PUBLIC_BUCKET_HOST> in all five addresses with the public host of your
   bucket.
5. Open n8n Form Trigger, copy its test address, send a prompt and a style, and check the picture.
6. Press Publish and share the production address. The form's path is flux4free.

How a run goes

1. n8n Form Trigger (2.1) serves a form titled flux.schnell image generator with a required text area, `Prompt to
   flux, and a dropdown, Style`. It waits for a response node to answer.
2. Route by style (Switch 3.1) sends the run to the Set node of the chosen style: Hyper-Surreal Escape, Post-Analog
   Glitchscape, AI Dystopia, Neon Fauvism or Vivid Pop Explosion. Its fallback output goes to a sixth node, None,
   which adds almost no style text.
3. The style's Set node writes stylePrompt, a long text of style words, and keeps the form's fields.
4. Call hugginface inference api (HTTP Request 4.2) posts Depict, your prompt and the style text as one prompt to
   the router. The reply holds the picture's link in images[0].url.
5. Download image fetches the link as a file. Upload image to S3 saves it as fg- plus the run's number, with
   .jpg.
6. Serve webpage (Respond to Webhook 1.1) answers with an HTML page. It shows the new picture, the style's name and
   four older pictures, found by counting back from the run's number.
7. A failed call or upload goes to Respond with error, which answers the form with a short message.

Change the look

• Styles: the five Set nodes, each with a stylePrompt, and the dropdown in n8n Form Trigger. A new style needs a
  Set node, a rule in Route by style and a connection to the call.
• Model: the address in Call hugginface inference api. Other routes take other field names, so check the provider's
  page for the model.
• The page: the HTML and its style block in Serve webpage.
• The form: its title, description and fields in n8n Form Trigger.
• The picture's name: fileName in Upload image to S3. Serve webpage reads the same name, so change both.

Good to know

• Mindket could not run this copy with a live Hugging Face token. The route and the shape of the reply follow Hugging
  Face's own client code, so send one test prompt before you share the form.
• Hugging Face's listing for FLUX.1 schnell, read on 12 November 2026, names fal.ai and WaveSpeed as its inference
  providers. The route in this copy goes to fal.ai.
• Anyone who has the production address can send prompts, and each picture is billed to your Hugging Face account. Turn on
  Basic Auth in n8n Form Trigger before you share it widely.
• The four older pictures count back from the run's number, so a run number used by another workflow, or by a failed
  run, leaves a broken picture there.
• Each style's text ends in flags like --ar 16:9, written for another image tool. The model reads them as words, so edit
  the Set nodes if you want a clean prompt.
• Each prompt goes to Hugging Face and its provider. The pictures go to your bucket and anyone with its address can see
  them.
• The file holds no credential and no key.