OpenAI chat examples: summary, German, SVG and quick replies

OpenAI chat examples: summary, German, SVG and quick replies

An n8n workflow with five small examples of calling OpenAI's chat models, all fed the same short sample text. It
summarises the text and translates the summary. It adds a system message and sends a list of messages through the HTTP
Request node. It also writes a prompt for a cover image, asks for SVG code and drafts quick email replies. A sixth branch,
switched off, transcribes an audio file.

What is inside

• workflow.json: the workflow as n8n exports it, twenty-three nodes: fourteen that do the work and nine sticky notes.
  It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The credential links are gone. Select your own OpenAI account.
• The original opened with a branch that used OpenAI's text completion and text edit calls on text-davinci-003. OpenAI's
  deprecations page, read on 12 November 2026, says that model and the /v1/edits endpoint shut down on 4 January 2024,
  and it names no edit replacement apart from chat completions. Mindket's copy drops those two nodes and the note that
  introduced them.
• The last step of the third example made four pictures with DALL-E 2. The same page says dall-e-2 and dall-e-3 were
  removed on 12 May 2026. That node is gone too, and the example now ends with the written picture prompt.
• Every chat node named gpt-3.5-turbo-0301, or named nothing and took the node's default, gpt-3.5-turbo. The first shut down
  on 13 September 2024. The second shut down on 23 October 2026, and OpenAI names gpt-5.6-terra as its replacement.
  All seven now name gpt-5.6-terra.
• The original also set a token limit and a temperature on several chat nodes. Mindket's copy drops both, so each call uses
  the model's own settings. The fifth example still asks for three replies.
• The example text was a transcript of a podcast episode. It is now a short text of our own about herbs in a window box.
• The sample audio path now reads YOUR_AUDIO_FILE.mp3.
• Three typos in the prompts, "wollowing", "followint" and "promt", are corrected.
• Nothing else.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• For the audio branch only: an audio file on the machine that runs n8n, in a folder n8n can read.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential. Select it in the six OpenAI nodes ChatGPT-ex1.1, ChatGPT-ex1.2, ChatGPT-ex2,
   ChatGPT-ex3.2, ChatGPT-ex4 and ChatGPT-ex. In ChatGPT-ex3.1, the HTTP Request node, choose the OpenAI
   credential type and select the same one.
3. Open the last node of a branch, such as ChatGPT-ex1.2, and execute it with its earlier nodes. Do one branch, then
   the next.
4. For the audio branch, set the path of your file in LoadMP3 and turn on LoadMP3 and Whisper-transcribe. Turn
   Text-example off so the transcript passes through. Read the note on whisper-1 below before you rely on this branch.

How a run goes

When clicking "Execute Workflow" starts the example branches. Each one reads the sample text from Text-example, or
carries its own.

1. Example 1: ChatGPT-ex1.1 writes a summary and ChatGPT-ex1.2 translates it to German. Only user messages are sent.
2. Example 2: ChatGPT-ex2 sends a system message that asks for five emojis at the end, then the user message.
3. Example 3: Code-ex3.1 builds the list of system and user messages. ChatGPT-ex3.1 sends it through the HTTP Request
   node, and ChatGPT-ex3.2 turns the answer into the image prompt.
4. Example 4: Set-ex4 holds a prompt that asks for HTML with an SVG. ChatGPT-ex4 answers and HTML-ex4 renders the code.
5. Example 5: ChatGPT-ex drafts three short replies to an email, five to eight words each.
6. Audio, switched off: LoadMP3 reads the file, and Whisper-transcribe sends it to the transcription call.

Change the look

• The sample text: the Code node Text-example.
• The model: the Model value in each OpenAI node and the model body parameter of ChatGPT-ex3.1.
• The instructions: the system messages in ChatGPT-ex2, Code-ex3.1, ChatGPT-ex3.2 and ChatGPT-ex.
• How many replies the fifth example drafts: Number of Completions in ChatGPT-ex.

Good to know

• Each branch is a separate call to OpenAI, and a sticky note in the file warns that the whole workflow is slow.
• Each OpenAI node here is n8n's older OpenAI node, which n8n hides from its node list. It still runs from this file.
• The third example's HTTP Request node sends only model and messages. The original also set a temperature, a token
  limit and a count of one, which are gone.
• The fifth example keeps its request for three replies. Check that your model accepts it, and set the count to 1 if it
  does not.
• whisper-1 shuts down on 26 February 2027, and OpenAI's deprecations page names gpt-live-transcribe or gpt-transcribe
  as the replacement.
• The text goes to OpenAI.
• The JSON holds no credential.