Generate research-backed infographics with GPT-4, kie.ai, and Gmail

Form input → AI web research & prompt optimization → kie.ai image generation → polling → email delivery

Who

Marketers, content creators, educators who need data-driven visuals fast

You get

Receive a research-backed infographic as email attachment after form submission; detailed error email on failure

You need

n8n · kie.ai · Gmail

Steps

1. Configure credentials and recipient

   After opening the template in n8n, configure credentials in this order: 1) Add your OpenAI API key in the Researcher node (OpenAI Chat Model); 2) Create a Header Auth credential with your kie.ai Bearer token and link it to both Generate Infographic and Check Job Status HTTP Request nodes; 3) Connect your Gmail OAuth2 account in both Email Image to User and Send Error Email nodes; 4) Change the sendTo field from '[email protected]' to the actual recipient address in both Gmail nodes.

2. Fill the form to trigger workflow

   Activate the workflow, then open the form URL provided by the On form submission node. Form fields: Main Headline (5-7 words), Key Data Points (topic/description), Target Audience (dropdown), Tone of Voice (dropdown), Visualization Options (dropdown), Art Style (dropdown), Color Palette (text input), Layout Structure (dropdown), Font Family (dropdown), Information Density (radio), Aspect Ratio (dropdown), Resolution (radio: 1K/2K/4K), Output Format (radio: png/jpg). All fields are required. Submission automatically triggers the workflow.

3. AI Agent researches topic and builds image prompt

   Form data flows into the Build Image Prompt node (AI Agent). This node calls the Researcher (GPT-4.1-mini with web search) to gather required data points, then follows system instructions to convert user requirements into a highly optimized diffusion-model prompt. The prompt emphasizes readability, composition, white space, vector style, and formats the headline as capitalized quoted text "HEADLINE".

   Execution prompt · Source text。Build Image Prompt node text parameter (AI Agent system instruction)

   Copy prompt

   =### ROLE
   You are a world-class Art Director and Prompt Engineer specializing in generative AI for information design. Your goal is to take user requirements and convert them into a highly optimized image generation prompt for a diffusion model.
   
   ### YOUR GOAL
   Create a prompt that results in a clean, legible, and professional infographic. You must prioritize "readability" and "composition" over artistic flair.
   
   ### INPUT DATA
   - Headline: {{ $json['Main Headline'] }}
   - Data Points: {{ $json['Key Data Points'] }}
   - Layout: {{ $json['Layout Structure'] }}
   - Visualization Types: {{ $json['Visualization Options'] }}
   - Tone of Voice: {{ $json['Tone of Voice'] }}
   - Style: {{ $json['Art Style'] }}
   - Information Density: {{ $json['Information Density'] }}
   - Font Family: {{ $json['Font Family'] }}
   - Colors: {{ $json['Color Palette'] }}
   
   ### TOOLS
   Use the "Researcher" model, which has access to the Internet, to gather the "Data Points" that are needed for the infographic.
   
   ### PROMPT STRUCTURE RULES
   1. **Subject & Layout:** Start by defining the image type strictly (e.g., "A vertical infographic vector illustration...").
   2. **Text Handling:** Modern AI models (like Flux or SD3) can render text if it is capitalized and wrapped in quotes. You MUST extract the "Headline" and place it in the prompt as: text "HEADLINE HERE". Do not try to fit all data points as literal text; summarize them as visual icons or short labels.
   3. **Style Descriptors:** Use keywords that enforce cleanliness: "white space," "margins," "grid layout," "vector graphics," "adobe illustrator style," "high resolution."
   4. **Negative Weighting (Implicit):** phrasing matters. Avoid "cluttered," "messy," "illegible."
   
   ### OUTPUT FORMAT
   Return ONLY the raw prompt string to be sent to the image generator. Do not add conversational filler or the URLs of the sources for "Data Points".
   
   ### DYNAMIC PROMPT TEMPLATE (Use this logic)
   "[Style] infographic featuring text ' [Headline] ' at the top. The layout is [Layout]. Preferred visualization types: [Visualization Types]. Tone of voice: [Tone of Voice]. Art style: [Style]. Information density: [Information Density]. Preferred font family [Font Family]. The design uses a [Colors] color scheme.
   Key visual elements include: [Convert 'Data Points' into visual metaphors, e.g., if data is 'costs down', describe a downward arrow or coin stack].
   High quality, professional graphic design, flat vector art, precise lines, extra white space for readability, typography focused."

4. Submit generation job to kie.ai and start polling

   The prompt from Build Image Prompt flows into the Generate Infographic node (HTTP Request), which POSTs to https://api.kie.ai/api/v1/jobs/createTask with model fixed as nano-banana-pro, plus user-selected aspect_ratio, resolution, and output_format from the form. On success, the workflow enters Polling Delay (15s) to wait 15 seconds, then Check Job Status queries https://api.kie.ai/api/v1/jobs/recordInfo. The Route by Job Status node branches on the state field: success → download image; waiting/generating → Increment Retry count and loop back to polling; fail → error handling. The Check Timeout node triggers timeout error handling when retryCount > 20.

5. Download image and email it, or send error email

   Success path: Download Generated Image parses resultUrls[0] from resultJson to fetch the image, and Email Image to User sends it as an attachment to the configured recipient. Error path has three branches: 1) Generation failure (state='fail'): Prepare Error Email extracts errorMessage and taskId, Send Error Email sends details; 2) Polling timeout (>20 retries): email explains timeout with original request info; 3) API error: email includes full error JSON. All error emails contain key original form fields for troubleshooting.

   Execution prompt · Source text。Prepare Error Email node jsCode parameter

   Copy prompt

   const input = $input.first().json;
   const formData = $('On form submission').item.json;
   let subject, body;
   
   if (input.data?.state === 'fail') {
     subject = 'Infographic generation failed';
     body = `Hi,\n\nThe infographic generation failed.\n\n**Error Details:**\n${input.data.errorMessage || 'Unknown error'}\n\n**Task ID:** ${input.data.taskId || input.taskId || 'unknown'}\n\n**Original Request:**\n- Headline: ${formData['Main Headline']}\n- Data Points: ${formData['Key Data Points']}\n- Style: ${formData['Art Style']}\n- Layout: ${formData['Layout Structure']}\n- Aspect Ratio: ${formData['Aspect Ratio']}\n- Resolution: ${formData['Resolution']}`;
   } else if (input.retryCount !== undefined) {
     subject = 'Infographic generation timed out';
     body = `Hi,\n\nThe infographic generation timed out after 20 polling attempts.\n\n**Task ID:** ${input.taskId || 'unknown'}\n\n**Original Request:**\n- Headline: ${formData['Main Headline']}\n- Data Points: ${formData['Key Data Points']}\n- Style: ${formData['Art Style']}\n- Layout: ${formData['Layout Structure']}\n- Aspect Ratio: ${formData['Aspect Ratio']}\n- Resolution: ${formData['Resolution']}\n\nPlease check the kie.ai dashboard for more details.`;
   } else {
     subject = 'Infographic API Error';
     body = `Hi,\n\nAn API error occurred during infographic generation.\n\n**Error Details:**\n${JSON.stringify(input, null, 2)}\n\n**Original Request:**\n- Headline: ${formData['Main Headline']}\n- Data Points: ${formData['Key Data Points']}\n- Style: ${formData['Art Style']}\n- Layout: ${formData['Layout Structure']}`;
   }
   
   return [{ json: { subject, body } }];

You are reproducing an n8n template: use GPT-4 with web search to research a topic, generate an infographic via kie.ai, and deliver results via Gmail. First confirm you have access to an n8n instance, an OpenAI API key, a kie.ai API key (Header Auth), and a Gmail OAuth2 credential. Then configure nodes step by step and replace the default recipient address. The main prompt (Build Image Prompt) is provided as the node's system instruction and requires no modification.