VPS Scraper Modfication to include AI Rewriting of Property Content

I have an existing scraper that imports property listings from a Vietnam real estate website into my own property portal. I require the following modifications: 1. AI Rewriting of Property Content The scraper currently translates Vietnamese titles and descriptions into English. I now want the translated content to be further processed by AI to create unique, natural-sounding English content. Requirements: * Rewrite property titles into unique English titles. * Rewrite property descriptions into unique, human-readable English descriptions. * Preserve all factual property information. * Avoid duplicate content. * Content should be suitable for SEO. * The system should run automatically as part of the scraping/import process. 2. Automated AI Image Processing Pipeline I want all imported property images processed automatically on the VPS before being stored. Required pipeline: 1. Download original image 2. Real-ESRGAN upscale (2x) 3. Small automatic crop (approximately 2–5%) 4. SDXL image-to-image enhancement * Denoising strength: approximately 0.25 * Preserve room layout and property accuracy * Improve lighting, colour balance, and overall image quality 5. Convert final image to AVIF format 6. Save processed image and use it within the property listing Objectives: * Create visually unique images * Improve image quality * Maintain accurate representation of the property * Reduce duplicate image issues * Generate efficient AVIF images for faster page loading Existing Environment * Existing scraper already operational * Linux VPS * Property listings imported automatically * Looking for a fully automated solution integrated into the current workflow Please Include * Recommended AI models * VPS/GPU requirements * Estimated processing time per image * Estimated monthly processing capacity * Previous experience with SDXL, Stable Diffusion, Real-ESRGAN, or similar image-processing pipelines Expected Processing Volume Current volume is approximately 54,000 images per month. A previous technical assessment estimated the following processing times for 1600px property photos: - Real-ESRGAN 2x Upscale: ~3 seconds - SDXL Image-to-Image: ~6 seconds - Crop & AVIF Conversion: ~1 second - Total: ~10 seconds per image Expected volume is approximately: - 1,800 images per day - 54,000 images per month A proposed deployment architecture is: - Existing scraper remains on the Contabo VPS - Image processing runs on a dedicated GPU service such as RunPod - Recommended GPU: RTX 4090 - Estimated GPU processing cost: approximately $100–300/month depending on image resolution, workflow optimizations, and GPU utilization The SDXL image-to-image stage is expected to account for the majority of processing time and cost. Please advise if you would recommend a different architecture, model, or optimization strategy for handling this volume while maintaining image quality and uniqueness.