feat: improve gemini prompt for intelligent handling of thin objects and text removal

backend/server.js

@@ -387,7 +387,13 @@ app.post('/api/sparkplug/generate-images', authenticate, async (req, res) => {
 
     // Run parallel generation for all 4 views
     const promises = views.map(async (view, index) => {
-      const fullPrompt = `${prompt}. ${view}, isolated on a pure white background, studio lighting.`;
+      const fullPrompt = `A clean, isolated product photography shot of: ${prompt}. 
+Focus ONLY on the product itself. 
+View: ${view}.
+CRITICAL INSTRUCTIONS:
+1. Pure white background.
+2. DO NOT include any infographics, extra text, labels, icons, or background scenery. Just the isolated object.
+3. If this object is very thin (like a sachet, pouch, or packet), the 'Left side view' and 'Right side view' must only show the thin edge profile of the object without the main front artwork.`;
       
       const payload = {
         contents: [