Modern Image Formats: AVIF and WebP

By Michael Samuel Naeem · May 10, 2026 · 6 min read

Our product catalog served 400KB JPEG hero images. Converting to AVIF at equivalent visual quality produced 120KB files — a 70% reduction across 2,000 product images. Total page weight on category pages dropped from 4.2MB to 1.4MB. LCP improved by 2.1 seconds on mobile without changing image dimensions or CDN infrastructure.

Format comparison

Picture element with fallbacks

<picture>
  <source srcset="/images/hero-800.avif 800w, /images/hero-1200.avif 1200w"
          sizes="(max-width: 768px) 100vw, 1200px" type="image/avif" />
  <source srcset="/images/hero-800.webp 800w, /images/hero-1200.webp 1200w"
          sizes="(max-width: 768px) 100vw, 1200px" type="image/webp" />
  <img src="/images/hero-1200.jpg"
       srcset="/images/hero-800.jpg 800w, /images/hero-1200.jpg 1200w"
       sizes="(max-width: 768px) 100vw, 1200px"
       width="1200" height="675" alt="Product hero" loading="lazy" decoding="async" />
</picture>

The browser picks the first source whose type it supports. JPEG is the universal fallback.

Responsive srcset

Serve appropriately sized images per viewport:

<img src="/photo-800.avif"
     srcset="/photo-400.avif 400w, /photo-800.avif 800w, /photo-1200.avif 1200w"
     sizes="(max-width: 600px) 100vw, (max-width: 1200px) 50vw, 600px"
     width="800" height="600" alt="Description" />

Build-time conversion

Convert during CI with sharp:

import sharp from 'sharp';

await sharp('src/images/hero.jpg')
  .avif({ quality: 50, effort: 4 })
  .toFile('dist/images/hero.avif');

await sharp('src/images/hero.jpg')
  .webp({ quality: 80 })
  .toFile('dist/images/hero.webp');

Generate all formats at build time. Serve from static hosting or CDN.

Quality tuning

Compare visually at 100% zoom on retina displays. Automated SSIM comparison tools help batch-tune quality settings.

CDN automatic format negotiation

https://cdn.example.com/cdn-cgi/image/format=auto,quality=80/photo.jpg

The CDN inspects the Accept header and returns the best supported format. No picture element needed.

Lazy loading and priority

Preload the LCP image with <link rel="preload">. Use loading="lazy" for below-the-fold images only. Never lazy-load the LCP candidate.

Checklist

1. Convert images to AVIF and WebP at build time
2. Use <picture> with format fallbacks
3. Add responsive srcset and sizes
4. Set explicit width and height to prevent CLS
5. Preload LCP image, lazy-load everything else
6. Tune quality per format

Art direction with picture

Serve different crops per viewport:

<picture>
  <source media="(max-width: 768px)" srcset="/hero-mobile.avif" />
  <source media="(min-width: 769px)" srcset="/hero-desktop.avif" />
  <img src="/hero-desktop.jpg" alt="Hero" width="1200" height="675" />
</picture>

Combine art direction with format fallbacks using nested picture elements or multiple source tags.

Encoding speed vs. quality

AVIF encoding is slower than WebP or JPEG. Pre-generate at build time — never encode on-the-fly in request path unless using a dedicated image CDN with caching.

Debugging checklist

When something doesn't work as documented, verify browser support with Can I use before assuming a polyfill bug. Check the Network tab for failed resource loads, incorrect MIME types, and missing CORS headers. Use the Console for CSP violations and Trusted Types errors that silently block operations.

Compare behavior in incognito mode to rule out extension interference. Test with cache disabled during development but validate with realistic caching in staging. Read the specification for edge cases the tutorial skipped — MDN examples cover happy paths, not every boundary condition.

If performance regresses after deployment, roll back first and investigate second. Keep a changelog of performance-related changes linked to metric dashboards. Future you will need to know why that preload tag exists before removing it during a refactor.

Screenshot and UI imagery

AVIF lossy compression blurs small text in marketing screenshots—use WebP lossless or PNG for UI captures with text. Photos and hero photography benefit most from AVIF.

Email clients

Most email clients ignore AVIF—serve JPEG in <img> for transactional email; AVIF for web only.

Image CDNs and origin shield

Transform at CDN edge on first request caches AVIF variant at PoP—origin serves single JPEG master. Configure quality rungs (q=45, q=55, q=65) and let CDN negotiate.

picture element cascade

<picture>
  <source type="image/avif" srcset="hero.avif" />
  <source type="image/webp" srcset="hero.webp" />
  <img src="hero.jpg" alt="..." width="1200" height="630" />
</picture>

Always include JPEG fallback for email clients and old Safari. CDN auto-negotiation simplifies markup but test Vary headers cache correctly.

AVIF for UI screenshots

Lossy AVIF blurs small text in product screenshots — use WebP lossless or PNG for UI captures; AVIF for photo heroes.

Practical follow-through (1)

Ship the smallest vertical slice first — one route, one widget, one index configuration — with rollback documented before expanding scope. Baseline the user-visible metric this work protects (latency, recall, conversion, task success rate) for seven days before change and seven days after in your largest market.

Compare canary p75 to control before full rollout. Exercise edge paths manually: refresh, back navigation, double-submit, offline mode, and keyboard-only flows. When assumptions change — traffic doubles, vendor upgrades, org restructure — revisit whether the original design still fits; quiet periods hide drift until the next incident.

Practical follow-through (2)

Ship the smallest vertical slice first — one route, one widget, one index configuration — with rollback documented before expanding scope. Baseline the user-visible metric this work protects (latency, recall, conversion, task success rate) for seven days before change and seven days after in your largest market.

Compare canary p75 to control before full rollout. Exercise edge paths manually: refresh, back navigation, double-submit, offline mode, and keyboard-only flows. When assumptions change — traffic doubles, vendor upgrades, org restructure — revisit whether the original design still fits; quiet periods hide drift until the next incident.

Practical follow-through (3)

Ship the smallest vertical slice first — one route, one widget, one index configuration — with rollback documented before expanding scope. Baseline the user-visible metric this work protects (latency, recall, conversion, task success rate) for seven days before change and seven days after in your largest market.

Compare canary p75 to control before full rollout. Exercise edge paths manually: refresh, back navigation, double-submit, offline mode, and keyboard-only flows. When assumptions change — traffic doubles, vendor upgrades, org restructure — revisit whether the original design still fits; quiet periods hide drift until the next incident.

Practical follow-through (4)

Ship the smallest vertical slice first — one route, one widget, one index configuration — with rollback documented before expanding scope. Baseline the user-visible metric this work protects (latency, recall, conversion, task success rate) for seven days before change and seven days after in your largest market.

Compare canary p75 to control before full rollout. Exercise edge paths manually: refresh, back navigation, double-submit, offline mode, and keyboard-only flows. When assumptions change — traffic doubles, vendor upgrades, org restructure — revisit whether the original design still fits; quiet periods hide drift until the next incident.

Practical follow-through (5)

Ship the smallest vertical slice first — one route, one widget, one index configuration — with rollback documented before expanding scope. Baseline the user-visible metric this work protects (latency, recall, conversion, task success rate) for seven days before change and seven days after in your largest market.

Compare canary p75 to control before full rollout. Exercise edge paths manually: refresh, back navigation, double-submit, offline mode, and keyboard-only flows. When assumptions change — traffic doubles, vendor upgrades, org restructure — revisit whether the original design still fits; quiet periods hide drift until the next incident.

Resources

• web.dev: Choose the right image format
• AVIF browser support (Can I use)
• MDN: picture element
• Squoosh image compression tool
• sharp image processing library