Image Processing Feature Specification Template

Start with JPEG, PNG, and WebP. These three cover 99% of web use cases. Add HEIC if your mobile user base is significant (iPhone default since iOS 11). Add AVIF if you are optimizing for best-in-class performance and your CDN supports it. Add SVG only if your use case requires vector graphics, and always sanitize for XSS. For technical decision frameworks as a PM, see the [Technical PM Handbook](/technical-pm-guide).

Pre-generate variants on upload for high-traffic images with predictable sizes (profile photos, product thumbnails). Use on-the-fly transformation (via Imgix, Cloudinary, or Cloudflare Image Resizing) for content with unpredictable display sizes or low traffic. The hybrid approach works well: pre-generate the 3-4 most common sizes and use on-the-fly for everything else.

Layer AI moderation with human review. Use an AI service (AWS Rekognition, Google SafeSearch) as the first pass with a confidence threshold. Images above the threshold are auto-rejected. Images in the uncertain range go to a human review queue. Images below the threshold are auto-approved. Track false positive and false negative rates and adjust thresholds monthly. Document your content policy clearly so moderators make consistent decisions. Use the [definition of done template](/templates/definition-of-done-template) to define moderation criteria.

For web performance, target under 200KB total image weight above the fold. Individual images should be under 100KB for inline content. Use WebP as the default format (25-35% smaller than JPEG at equivalent quality). Set JPEG quality to 80-85% and WebP quality to 75-80%. Lazy load images below the fold. These targets align with Google's Core Web Vitals thresholds for LCP.

Queue uploads locally and sync when connectivity returns. Show a local preview (from the device's file system) immediately. Display upload progress with estimated time. Allow users to cancel queued uploads. For very large files on slow connections, consider chunked upload (tus protocol) with resume capability. Define the maximum queue size and what happens when it fills up. Track your upload metrics in a [KPI dashboard](/templates/kpi-dashboard-template) to identify bandwidth-related failures.