Image Optimization Expert

Default Workflow

Inspect the real surface before prescribing changes: rendered HTML, framework image components, network waterfall, build output, CDN URLs, and asset inventory.
Identify the page role of each important image: LCP/hero, above-fold content, product/gallery, thumbnail/list item, CSS background, decorative, user-uploaded, or source/master asset.
Fix the highest-impact path first: LCP delivery, oversized images, missing dimensions, wrong format, missing responsive candidates, unnecessary client-side resizing, or CDN bypass.
Keep source quality high and delivery variants cheap: preserve master assets, transform at the CDN/build edge, and avoid destructive re-encoding loops.
Verify with evidence: rendered markup, response headers/content type, decoded/displayed size, Lighthouse/WebPageTest/DevTools, and visual inspection for artifacts.

When a repo/app is available, make the patch instead of only giving advice. Prefer local framework conventions over generic snippets.

Fast Audit Checklist

LCP: Hero image is not lazy loaded, has fetchpriority="high" or framework priority, is preloaded when appropriate, comes from a fast origin/CDN, and is right-sized for the rendered slot.
CLS: Every meaningful image has intrinsic width and height or a stable CSS aspect-ratio.
Responsive delivery: Width-based srcset uses realistic layout sizes; CSS/background images have equivalent responsive handling.
Formats: Use negotiated/modern formats (format=optimal, AVIF/WebP, or framework auto format) with safe fallbacks.
Compression: Quality is content-aware, not universally maxed out; product/face/detail images get higher quality than thumbnails/backgrounds.
Loading: Below-fold images are lazy and async-decoded; above-fold images are eager and not hidden behind slow JS.
CDN/cache: Image URLs are cacheable and transformation params are stable; source files are not repeatedly transformed by multiple layers.
Accessibility/SEO: Informative images have useful alt text; decorative images use empty alt; product assets can preserve metadata in CMS/CDN records even if delivery strips EXIF.

Use The Audit Script

For HTML files or public URLs, run the bundled no-dependency audit before and after changes. The script lives in this skill's own scripts/ directory — resolve it from the skill's base directory (shown when this skill loads), not the target project:

node <skill-base-dir>/scripts/audit-images.mjs ./dist/index.html
node <skill-base-dir>/scripts/audit-images.mjs --head https://example.com/
node <skill-base-dir>/scripts/audit-images.mjs --json https://example.com/ > image-audit.json

The script checks image markup, srcset/sizes, preload hints, likely LCP mistakes, oversized CDN widths, CSS background images, known image CDNs (Sirv, Cloudinary, imgix, Cloudflare, Next.js, and others), and optional HTTP headers/content length. It reads the server HTML only; images that client-side JS injects do not appear, so compare with DevTools for SPAs.

Decision Matrix

Situation	Preferred Action
Static/project-owned images	Generate responsive variants at build time or move masters to an image CDN.
User-uploaded or product catalog images	Use an image CDN/API pipeline; do not commit generated variants into the app repo.
Existing Sirv account or Sirv URLs	Use Sirv dynamic imaging/profile/API workflows; read sirv-workflows.md.
Product gallery, zoom, 360 spin, video/model viewer, or smart gallery	Use `../sirv-media-viewer/SKILL.md` before hand-building gallery behavior.
Next.js app	Prefer `next/image`; use a custom loader for external image CDNs when the CDN should transform. In Next.js 16+, `priority` is deprecated: use `fetchPriority="high"` (or `preload` for exactly one LCP image), and add every `quality` value to `images.qualities`.
CSS background hero	Consider replacing with semantic `<img>`/`picture`; if it must remain CSS, use `image-set()` with `type()` and preload carefully.
Image quality problem	Compare candidate qualities visually and with SSIM/VMAF/Butteraugli where possible.
Need background removal, upscaling, product lifestyle, or alt text at scale	Use `sirv-ai-studio` when available; keep delivery concerns in Sirv/CDN workflow.

When To Read References

Sirv implementation workflows: CDN migration, dynamic URLs, profiles, REST inventory/upload/search, Next.js loader, verification: sirv-workflows.md
Format details: browser support, encoding options, SVG security, HDR/wide gamut: formats.md
Compression and quality: quality settings, metadata, batch processing, SSIM/VMAF: optimization.md
Responsive images: srcset, sizes, art direction, priorities, container queries, backgrounds: responsive.md
Performance: LCP/CLS/INP, placeholders, preload, budgets, measurement: performance.md
Tools and services: image CDN URL syntax, Next.js/Astro/Vite/Eleventy integration, encoder flags, Sharp: tools.md

Also load sibling skills when the task crosses into their specialty:

../video-optimization/SKILL.md for hero/background video, GIF-to-video, streaming, and video embeds.
../sirv-dynamic-imaging/SKILL.md for Sirv URL parameters, profiles, crops, overlays, and caching.
../sirv-api/SKILL.md for Sirv auth, file upload/fetch, search, metadata, jobs, usage limits.
../sirv-media-viewer/SKILL.md for Sirv Media Viewer markup, JS loading, smart galleries, zoom/spin/video/model/PDF galleries, viewer API/events, accessibility, CSP, and performance.
../sirv-ai-studio/SKILL.md for AI processing, MCP tools, background removal, upscaling, generation, product workflows, and alt text.

Default Recommendations

Use Case	Format/Delivery Default
Photos and product images	AVIF/WebP via auto negotiation; JPEG fallback when needed.
Graphics/logos/icons	SVG when trusted and simple; otherwise WebP/PNG as needed.
Screenshots/text-heavy raster images	WebP lossless/near-lossless or optimized PNG.
Short animations	Animated WebP or video; avoid large GIFs.
High-value hero/product detail	Slightly higher quality, explicit dimensions, responsive candidates, CDN edge delivery.
Thumbnails/list media	Aggressive resizing, lower quality, lazy loading, stable aspect ratio.

Quality starting points:

Format	Starting Quality
JPEG	75-85, higher for product/detail
WebP	75-85 lossy; lossless for sharp graphics
AVIF	60-75 for most photos
PNG	Lossless, then optimize or convert when safe

Common Mistakes To Catch

Lazy-loading the LCP image, or hiding it behind JS (data-src swaps, client-only rendering).
Missing dimensions/aspect ratio.
Serving a 2-4K image into a small card or mobile slot.
Writing sizes="100vw" for images that render in a 33-50vw grid.
Combining framework optimization and CDN transformations in ways that double-compress or block caching.
Hard-coding AVIF/WebP without a fallback or content negotiation path.
Using PNG for photos or JPEG for transparency/sharp text.
Treating alt text, filename, CDN metadata, and visible captions as interchangeable.
Fixing source assets but failing to verify rendered output and response headers.