Limits

edgesharp inherits the limits of its host: Cloudflare Workers + R2. The defaults are generous for typical web traffic (100 KB–10 MB JPEGs, 4K maximum, browser <Image> use), but raw DSLR files and very large dimensions are outside the envelope.

What scales

• Storage: R2 has no practical per-bucket cap. Stored output costs $0.015/GB-month; egress is free.
• Concurrency: Workers run per-request and stateless on the global edge. No fixed connection cap; CF spreads load across data centers.
• Cache hit rate: repeat requests for the same (url, width, quality, format) never re-transform; they hit Cache API or R2 in single-digit ms.

What doesn’t

Limit	Source	Where it bites
Workers isolate memory	128 MB	RGBA decode buffer = width × height × 4. A 5500×5500 image → 121 MB → out of memory. Practical ceiling: source dimensions ~4000×4000.
Worker request body	100 MB on Paid	Source images larger than this get rejected before edgesharp can decode them.
Worker CPU time per request	30 s on the Standard usage model (Paid default)	A 4K JPEG → AVIF transform is ~1–2 s CPU. Standard is the default for new Paid Workers, so cold transforms fit.
MAX_WIDTH constant	3840 (Next.js default)	Output width caps here regardless of source. Edit MAX_WIDTH in src/worker.ts if you need larger outputs.
MAX_URL_LENGTH	3072 chars	Source URL parameter cap. Long signed-URL inputs may need bumping.

Workers Paid is required

edgesharp needs Workers Paid ($5/month per Cloudflare account). Workers Free is not supported: the 10 ms CPU/request cap is too tight for cold transforms (a 4K JPEG → AVIF is ~1–2 seconds CPU), and Durable Objects, required for the warm WASM pool, are Paid-only.

Practical sweet spot

	Comfortable	Stretches	Fails
Source file size	< 10 MB	10–50 MB	> 100 MB (Workers Paid request body cap)
Source dimensions	< 4000×4000	4000–5000 px on a side	> 5500 px (memory)
Output width	up to 3840	n/a (capped)	n/a
Output format	any of jpeg/png/webp/avif	n/a	n/a
Cold transforms / sec	~10	~50 (Standard usage model)	unbounded with concurrent isolates, but bills accumulate

What to do when you hit a limit

• Source file too large: pre-process at the origin. Most static-site generators downsize originals before publishing; do that, and edgesharp takes a reasonable input.
• Source dimensions too large: same as above, or use Cloudflare Images for the initial downsize. CF Images has higher resource limits as a managed service.
• CPU exceeded: Standard usage model (30 s CPU/request) is the default for new Paid Workers, so a fresh deploy from this repo is already in the right place. If you cloned the wrangler config from a much older Worker that was on Bundled (50 ms), the simplest fix is to redeploy as a new Worker.
• Memory exceeded: almost always means the source is bigger than ~4000×4000. Either downsize at origin or set IMAGE_BACKEND: cf-images for that traffic and let Cloudflare Images handle it.

What edgesharp doesn’t try to be

• A general image-transformation service (cropping, watermarking, focal-point detection, that’s Cloudflare Images).
• A photo-pipeline backend (RAW processing, color-managed exports, that’s Sharp/libvips on a real server).
• A video transformer (use Stream).

It’s specifically a <Image>-loader replacement that handles the formats Next.js’s default loader handles, on Cloudflare’s runtime instead of Vercel’s.