Auto-generate story short videos with n8n, GPT-4o-mini, Midjourney, Kling, and Creatomate

Input character and scene params → GPT-4o-mini generates a 3-part story prompt → Midjourney images → Kling image-to-video → Creatomate final composition with subtitles

Who

Content creators, social media bloggers, short-form video makers, educational creators

You get

One-click generation of a 3-part story short video with subtitles for rapid creative validation

You need

n8n · PiAPI · Midjourney · Kling · Creatomate

Steps

1. Configure basic parameters and API credentials

   Open the Basic Params node in the n8n workflow and enter your PiAPI x-api-key. This node contains three input fields: style (visual style, e.g. --s 500 --niji 6), character (character description), and situational_keywords (scene atmosphere keywords). These parameters feed into the GPT-4o-mini prompt generation.

   Setup guidance · Source text。n8n Sticky Note / Basic Params node configuration note

   Copy prompt

   ## Basic Setting Instruction
   Fill in **Basic Params** based on user's need. 
   The node concludes three section: style, character and situation_keyword.
   GPT-4o-mini will generate image description based on basic params mentioned above.

2. Generate a 3-part story prompt

   Run the GPT-4o-mini Generate Image Scenario Prompt node. It sends a request to PiAPI's chat completions endpoint with system prompt 'You are a helpful assistant.' and a user prompt that asks for a short story based on character, style, and situational_keywords, divided into three sections, returning strict JSON: {'title':'','prompt1':'','prompt2':'','prompt3':''}.

   Execution prompt · Source text。GPT-4o-mini Generate Image Scenario Prompt node system message

   Copy prompt

   You are a helpful assistant.

   Execution prompt · Source text。GPT-4o-mini Generate Image Scenario Prompt node user message

   Copy prompt

   Character is {{ $json.character }}, style is {{ $json.style }}, context keywords are {{ $json.situational_keywords }}, please generate a short story containing relevant elements and styles based on Character and context keywords. It should be concise and clear and divided into three sections. The return format is {'title':'','prompt1':'','prompt2':'','prompt3':''}, directly return json data format, not markdo2t include unusual characters such as hyphens,The key and value are enclosed in double quotes, but if you need to use quotes in the value string, please use single quotes.

3. Generate three Midjourney images and poll for completion

   Run the three Midjourney Generator nodes (First/Second/Third Image) in sequence. Each sends a POST to https://api.piapi.ai/api/v1/task with model=midjourney, task_type=imagine, aspect_ratio=2:3, process_mode=turbo. The prompt combines character + the corresponding prompt (prompt1/2/3) + style. Then poll via Wait → Get Task → Verify Status → Switch until status=completed, then retrieve temporary_image_urls.

   Execution prompt · Source text。Midjourney Generator of the First Image node jsonBody (prompt1; prompt2/3 share same structure with different variable names)

   Copy prompt

   ={
     "model": "midjourney",
     "task_type": "imagine",
     "input": {
       "prompt": "{{ $('Basic Params').item.json.character }},{{ $json.prompts.prompt1 }},{{ $('Basic Params').item.json.style }}",
       "aspect_ratio": "2:3",
       "process_mode": "turbo",
       "skip_prompt_check": false
     }
   }

4. Generate three Kling videos and poll for completion

   Use the three images to call the Kling video_generation endpoint (https://api.piapi.ai/api/v1/task) with model=kling, version=1.6, mode=pro, image_url set to the corresponding image URL, and prompt fixed as 'Natural swing'. Then poll via Wait → Get Task → Verify → Switch until each video's video_url is retrieved.

   Execution prompt · Source text。Generate the First Video node jsonBody (image_output1; 2/3 share same structure)

   Copy prompt

   ={
       "model": "kling",
       "task_type": "video_generation",
       "input": {
           "version": "1.6",
           "mode": "pro",
           "image_url": "{{ $json.image_output1 }}",
           "prompt": "Natural swing"
       }
   }

5. Compose final video in Creatomate

   Pre-create a video template in Creatomate and note its template_id. In the Final Video Combination node, send a POST to https://api.creatomate.com/v1/renders with template_id and modifications: map the three Kling videos to Video-1/2/3.source and the title to Text-1.text. Headers must include Authorization: Bearer <Creatomate_API_Key> and Content-Type: application/json.

   Execution prompt · Source text。Final Video Combination node jsonBody

   Copy prompt

   ={
     "template_id": "c10c62b6-d515-4f36-a730-f4646d1b7ee2",
   "modifications": {
       "Video-1.source": "{{ $('output1').item.json.video1 }}",
       "Video-2.source": "{{ $('output2').item.json.video2 }}",
       "Video-3.source": "{{ $('Get the Third Video URL').item.json.video3 }}",
       "Text-1.text": "{{ $('Get Prompt').item.json.prompts.title }}"
     }
   }

This template is a setup starting reference. Import workflow.json into n8n first, confirm PiAPI and Creatomate API credentials are filled in the corresponding node Headers/Body. Then configure Basic Params step by step, verify GPT-4o-mini prompt output, check that Midjourney and Kling polling nodes trigger correctly, and finally create a template in Creatomate and replace the template_id. It is recommended to prototype with Creatomate basic assets before connecting the full workflow.