Image Optimizer

A fast, self-hostable web app that turns one image into every size and format you actually need — web formats, social media presets, and a complete favicon pack — and hands them back as a single ZIP.

Drop your images in the browser, tick the targets you want, and download the optimized results. No accounts, no uploads kept on disk, no setup beyond running one Docker container.

Why

Preparing images for the web is repetitive: resize for Instagram, crop for an Open Graph card, generate a dozen favicon files, re-encode to WebP/AVIF for performance… usually across several tools. This does all of it in one drag-and-drop step, runs entirely on your own server, and never writes your images to disk.

Features

• Many targets at once — pick any combination of presets; each image is optimized for every selected target in parallel.
• Modern + classic formats — outputs WebP, AVIF, JPEG (progressive), and PNG.
• Wide input support — JPEG, PNG, WebP, AVIF, and iPhone HEIC/HEIF.
• Drop-in favicon pack — one click produces favicon.ico, all the PNG sizes, apple-touch-icon, site.webmanifest, and a ready-to-paste HTML snippet.
• Social presets — Instagram, LinkedIn, X, Facebook, Pinterest, Open Graph, plus email/web banners.
• Live progress — per-target progress streamed over Server-Sent Events.
• Privacy by design — images are processed in memory only and deleted right after download (or after a short timeout); nothing is stored.
• Single container — the Svelte UI is embedded into the Go binary, so the whole app is one small Docker image with no external services.

How it works

1. Add images — drag & drop or browse (JPEG, PNG, WebP, AVIF, HEIC).
2. Pick targets — choose one or more presets.
3. Optimize — each image is resized/re-encoded for every target; watch live progress.
4. Download — get a single ZIP with every optimized variant, named by preset.

Quick start

Run the prebuilt-style image locally (only Docker required):

docker build -t image-optimizer .
docker run -p 3000:3000 image-optimizer

Then open http://localhost:3000 and start dropping images. See Production image for deployment details and Development for the hot-reload setup.

Tech stack

• Go + Fiber v3 — single compiled binary, low memory footprint
• govips / libvips — the image processing engine
• Svelte + Vite — zero-runtime SPA, embedded via //go:embed
• Single Docker container — Fiber serves both the API and the SPA

Project layout

.
├── main.go # Fiber app, config, graceful shutdown, embeds frontend/dist
├── config/ # env-var configuration (port, limits, workers, job TTL)
├── handlers/ # HTTP handlers: health, upload, progress (SSE), download
├── processor/ # govips pipeline, worker semaphore, presets
├── frontend/ # Svelte + Vite app; dist/ is the go:embed target
├── Dockerfile # 3-stage build: Vite → Go (cgo+libvips) → debian-slim
└── docker-compose.yml # local dev with hot-reload

Note on frontend/dist: a small stub index.html is committed so go build works locally without first running Vite. The Docker frontend stage overwrites it with the real Vite build.

Supported formats & presets

Input: JPEG, PNG, WebP, AVIF, and HEIC/HEIF (iPhone photos). HEIC/HEIF are decoded via libvips' heifload (libheif/libde265, bundled in the Docker images).

The convert_* presets are the "just turn this into a usable file" path: faithful, high-quality format conversion at the original size (no crop). Handy for turning an iPhone HEIC straight into JPEG/PNG/WebP/AVIF without any other tool.

Output presets (registry: processor/preset.go):

Preset	Format	Dimensions	Notes
convert_jpeg	JPEG	original	quality 92, progressive — faithful conversion
convert_png	PNG	original	compression 6 (lossless)
convert_webp	WebP	original	quality 90 — faithful conversion
convert_avif	AVIF	original	quality 80, effort 4 — faithful conversion
compress_best	source	original	re-encode only — near-lossless, keeps the source format
compress_balanced	source	original	re-encode only — strong shrink, great quality
compress_max	source	original	re-encode only — aggressive (PNG uses a lossy palette when it helps)
website_webp	WebP	original	quality 80 (web-optimized)
website_avif	AVIF	original	quality 60, effort 4 (web-optimized)
jpeg_original	JPEG	original	quality 80, progressive
png_original	PNG	original	compression 6
instagram_square	JPEG	1080×1080	progressive
instagram_portrait	JPEG	1080×1350	progressive
instagram_story	JPEG	1080×1920	progressive
linkedin	JPEG	1200×627	progressive
twitter	JPEG	1200×675	progressive
facebook_post	JPEG	1200×630	progressive
pinterest_pin	JPEG	1000×1500	progressive
og_image	PNG	1200×630	compression 6
favicon	—	—	favicon pack (see below)
thumbnail	PNG	400×400	compression 6
email_header	JPEG	600×200	progressive
web_banner	JPEG	1920×480	progressive

Fixed-size presets center-crop to the target dimensions; *_original presets keep the source dimensions and just re-encode + strip metadata.

The compress_* presets are the "just make this smaller for the web" path: they keep the source format (JPEG→JPEG, PNG→PNG, WebP→WebP; HEIC→JPEG, SVG→PNG) and the original dimensions, tuning only the encoding across three honest tiers (best / balanced / max savings). PNG savings are modest — it has no quality knob — so only compress_max tries a lossy palette, and only keeps it when it actually comes out smaller.

Favicon pack

The favicon preset is a multi-file output: instead of one image it generates a complete drop-in icon set, bundled under a favicon/ folder in the ZIP:

• favicon.ico (multi-size 16/32/48, hand-built ICO container)
• favicon-16x16.png, favicon-32x32.png, favicon-48x48.png
• apple-touch-icon.png (180×180)
• android-chrome-192x192.png, android-chrome-512x512.png
• site.webmanifest and a README.txt with the exact <head><link> snippet

The pack is generated from a center-cropped square master. The multi-file plumbing lives in Preset.Kind / Result.Files (processor/preset.go, processor/favicon_vips.go, processor/ico.go).

Development

Hot-reload for both backend (Air) and frontend (Vite dev server):

docker compose up

• Backend: http://localhost:3000 (/health returns {"status":"ok"})
• Frontend dev server: http://localhost:5173

Production image

The single self-contained image (Vite build embedded into the Go binary):

docker build -t image-optimizer .
docker run -p 3000:3000 image-optimizer
curl http://localhost:3000/health # -> 200 {"status":"ok"}

The final image contains only the Go binary plus the libvips runtime libs (libvips42) on debian:bookworm-slim, and runs as a non-root user.

Health check

The image ships a Docker HEALTHCHECK that probes GET /health. The binary self-probes via its -healthcheck flag, so no curl/wget is needed in the minimal runtime image:

docker run -d -p 3000:3000 image-optimizer
docker ps # STATUS shows "healthy" once the start period elapses

Graceful shutdown

On SIGINT/SIGTERM the server stops accepting new connections, drains in-flight jobs (so their SSE clients receive a terminal event and the ZIP stays briefly downloadable), tears down libvips, then exits. Each phase is bounded by a 30s timeout so shutdown never hangs.

Job lifetime

Job state lives in memory only — there is no disk temp storage. A job is freed when its ZIP is downloaded, or by a background reaper after JOB_TTL_MINUTES (default 10), whichever comes first. This bounds memory for jobs that are never downloaded.

Environment variables

All configuration is via environment variables, read once at startup (the resolved values are logged). Invalid or non-positive numeric values fall back to the default rather than failing startup.

Variable	Default	Description
PORT	3000	TCP port the HTTP server listens on.
MAX_FILE_SIZE_MB	50	Per-file upload cap. Larger files are rejected with 400.
WORKER_COUNT	number of CPUs	Max concurrent libvips pipelines (the real concurrency limit).
JOB_TTL_MINUTES	10	How long a job's in-memory state is retained before the reaper frees it.

The whole-request multipart body limit is derived from MAX_FILE_SIZE_MB plus headroom for multiple files and multipart boundaries.

Requirements

• Go 1.26+ · Node 22+ · Docker
• libvips ≥ 8.14 (provided inside the Docker images; install locally only if building the Go binary outside Docker)

License

Released into the public domain under The Unlicense — do whatever you want with it.