Build Consistent AI Character Images with n8n + Google Nano Banana + Kie.ai Upscaling

Manual trigger → create timestamped Google Drive folder and tracking sheet → GPT-4 generates scene prompt → Kie.ai NanoBanana creates image → poll status → 4× upscale with face enhancement → archive to Drive and update Sheets

Who

Content creators and automation engineers who need consistent character visuals for social media or marketing

You get

One-click generation of style-consistent AI character images with automatic 4× upscaling, timestamped Google Drive archiving, and progress logged in Google Sheets

You need

n8n · google-drive · google-sheets

Steps

1. Project setup: create folder and tracking sheet

   After importing the template into n8n, verify credential configuration. The workflow starts with a manual trigger node (When clicking 'Execute workflow'). The Folder Name node uses JavaScript to generate a timestamped folder name (format: Post HH:MM AM/PM DD Month YYYY). Then Create folder creates that folder under a specified Google Drive parent; Create spreadsheet creates a Google Sheet with the same name; Move Sheet moves the sheet into the new folder; Edit Fields initializes Prompt, Temp Image Url, and Status fields; Add Fields appends the first row to the sheet.

2. Generate scene prompt

   The Story Creator Agent node is a Basic LLM Chain connected to an OpenAI Chat Model (default gpt-4.1-mini) and a Structured Output Parser. The system message defines a unified character portrait style template: 'A highly realistic modern cinematic portrait of the same girl, always facing the camera...' with Styling Keywords (Photorealism, cinematic photography, etc.). The LLM generates a scene-specific prompt based on this template, constrained by Json parser4 to the format {"prompt": "..."}. The Code1 node cleans escapes and normalizes whitespace to extract plain text. Story Status Update writes the generated prompt and status 'prompt created' back to Google Sheets row 2.

   Execution prompt · Source text。Story Creator Agent node messages.messageValues[0].message field; serves as LLM Chain system/style instruction

   Copy prompt

   A highly realistic modern cinematic portrait of the same girl, always facing the camera with a natural, expressive look. She is in [location/setting] wearing [modern, stylish outfit/clothing style], engaged in [action/mood]. Lighting and atmosphere are contemporary and cinematic, matching the environment with sleek tones and realistic detail. Ultra-detailed skin, expressive eyes, lifelike textures, cinematic depth of field, modern color grading, immersive realism.
   
   Styling Keywords (keep for all scenes):
   
   Photorealism, cinematic photography, modern fashion styling, expressive emotion, natural face, immersive lighting, dynamic depth, editorial aesthetic, storytelling tone.

3. Call Kie.ai to generate image

   The Create Task (no callback) node sends a POST request to https://api.kie.ai/api/v1/jobs/createTask. The body sets model to 'google/nano-banana-edit', with input containing the previous step's prompt and an array of 5 reference image URLs (image_urls), output_format: 'png', and image_size: 'auto'. The header must include Authorization: Bearer YOUR_TOKEN_HERE. Then Set TaskId extracts data.taskId from the response for later polling.

   Execution prompt · Source text。Create Task (no callback) node jsonBody field; replace image_urls with actual reference images

   Copy prompt

   ={
     "model": "google/nano-banana-edit",
     "input": {
       "prompt": "{{ $json.Prompt }}",
       "image_urls": [
                 "https://i.ibb.co/q332LXHf/Whats-App-Image-2025-09-11-at-15-54-09.jpg",
                 "https://i.ibb.co/rPRzBdK/Whats-App-Image-2025-09-11-at-15-54-11-1.jpg",
                 "https://i.ibb.co/Zz9HPfZ9/Whats-App-Image-2025-09-11-at-15-54-12-1.jpg",
                 "https://i.ibb.co/tp7rhyK7/Whats-App-Image-2025-09-11-at-15-54-12-2.jpg",
                 "https://i.ibb.co/N2wDdJpJ/Whats-App-Image-2025-09-11-at-15-54-12-3.jpg"
       ],
       "output_format": "png",
       "image_size": "auto"
     }
   }

4. Poll task status and extract results

   The Wait 1 Min node waits one minute, then Poll Task sends a GET request to https://api.kie.ai/api/v1/jobs/recordInfo with query parameter taskId from Set TaskId. The Switch node branches based on data.state and data.resultJson: if state='fail' → fail branch; if resultJson is empty → return to Wait 1 Min; if resultJson is non-empty → success branch. On success, Get ResultUrls uses JavaScript to parse resultJson and extract the resultUrls array.

   Execution prompt · Source text。Get ResultUrls node jsCode field; extracts image URLs from polling response

   Copy prompt

   // n8n Function node
   const items = $input.all();
   const out = [];
   
   for (const item of items) {
     // Your HTTP node sometimes returns an array at top-level
     const payload = item.json;
     const first = Array.isArray(payload) ? payload[0] : payload;
   
     // Get the stringified resultJson safely
     const resultJsonStr = first?.data?.resultJson ?? '';
   
     // Parse and extract
     let resultUrls = [];
     try {
       const parsed = JSON.parse(resultJsonStr);
       resultUrls = parsed?.resultUrls ?? [];
     } catch (e) {
       // leave resultUrls as []
     }
   
     out.push({ json: { resultUrls } });
   }
   
   return out;

5. 4× upscale with face enhancement

   The resultUrls[0] from Get ResultUrls feeds into the Upscale Image node. This node POSTs to https://api.kie.ai/api/v1/jobs/createTask with model set to 'nano-banana-upscale', input containing image: resultUrls[0], scale: 4, and face_enhance: true. Then Set TaskId for upscale extracts the new taskId, Wait 1 Min1 waits, Get Upscaled Image polls, Switch1 evaluates state, and Get ResultUrls from UpScale parses the final result URL.

   Execution prompt · Source text。Upscale Image node jsonBody field

   Copy prompt

   ={
     "model": "nano-banana-upscale",
     "input": {
       "image": "{{ $json.resultUrls[0] }}",
       "scale": 4,
       "face_enhance": true
     }
   }

   Execution prompt · Source text。Get ResultUrls from UpScale node jsCode field

   Copy prompt

   // Build a new array of JSON objects with only resultUrls
   return $input.all().map(item => {
     const result = JSON.parse(item.json.data.resultJson);
     return {
       resultUrls: result.resultUrls
     };
   });

6. Archive to Google Drive

   The Get Binary node downloads the image binary using resultUrls as the URL. Upload file uploads the file to Google Drive (defaulting to a hardcoded 'prompt' folder, with filename set to the upscale taskId). Move Images moves the uploaded file into the timestamped folder created in step 1.

7. Update tracking sheet status

   The Update Image Status node updates Google Sheets row 2, setting Status to 'image created' and Temp Image Url to resultUrls[0] from Get Binary. The workflow is now complete; view the final image in the Google Drive timestamped folder and the full status history in Sheets.

You are reproducing an n8n workflow template that uses Kie.ai's NanoBanana model to generate consistent character images and automatically upscale them 4×. First confirm you have: Kie.ai API Token, OpenAI API Key, Google Drive OAuth2, and Google Sheets OAuth2 credentials. Then import the template, replace the hardcoded Google Drive folder ID and 5 reference image URLs, and execute the manual trigger.