SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillstinify-images

tinify-images

Design & creation

View repository

Updated 10-16 09:35

About this skill

Tinify image tools

Tinify.dev processes images server-side and reports honest results: compression never returns a file larger than the input (you get the original plus optimized: false instead), while resize and conversion report size changes plainly because they can legitimately grow a file.

Limits that shape your plan: 40 MB and 50 megapixels per image, results expire after two hours, supported formats are AVIF/WebP/JPEG/PNG (GIF in the web tools and anonymous endpoints). Free API tier: 500 successful operations per month; failed calls are not billed.

Pick the right path

  1. Tinify MCP tools available (tool names like compress_image, resize_image, crop_image, convert_image, get_usage from an MCP server named tinify): use them. They validate paths, refuse to overwrite the source unless asked, write <name>.min.<ext> next to the input, and return byte counts. Check get_usage first when running a large batch so you do not burn through the monthly quota mid-job.
  2. No MCP but TINIFY_API_KEY is set (key format tnf_live_... or tnf_test_...): call the REST API directly (below) or use the CLI: npx -y @tinify-dev/client compress ./img.png --out ./out.
  3. Neither: for one-off files point the user at https://tinify.dev (free web tools, no account); for automation tell them keys are free at https://tinify.dev/developers - do not invent a key or skip authentication.

REST API essentials

Base: https://api.tinify.dev/api/v1 - Bearer auth - Idempotency-Key header required (any UUID; reuse it when retrying the same logical request).

curl -X POST https://api.tinify.dev/api/v1/images/compress \
  -H "Authorization: Bearer $TINIFY_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -F "file=@photo.jpg" \
  -F "quality_mode=balanced"          # balanced | best_quality | lossless

Other operations, same shape:

  • /images/resize - width, height, or scale; keep_aspect_ratio=true
  • /images/crop - x, y, width, height (pixels)
  • /images/convert - format=avif|webp|jpeg|png (rejects same-format with 422)
  • compress also accepts target_size_bytes (>= 10240, < input size; check target_size_achieved in the response - false means best effort)
  • GET /usage - plan, used, remaining

Responses wrap payload in data with a top-level request_id. Download the result promptly from data.download_url (relative to the API origin, expires in two hours). Errors use a stable envelope: {"error": {"code", "message"}, "request_id"} - retry only on 429/503, honoring Retry-After.

Working rules

  • Never silently overwrite the user's source file; write alongside it unless they explicitly ask for in-place replacement.
  • Report per-file before/after bytes and the total saved. If optimized is false, say so - do not present the original as a win.
  • Lossless mode is not available for JPEG, and cannot combine with target_size_bytes.
  • For batches, process a few files at a time rather than all at once, and fail the batch loudly on auth/quota errors instead of skipping files quietly.
  • Conversion to a "smaller" format can produce a bigger file (e.g. small PNG icons to WebP). Compare bytes and keep whichever the user actually asked for, stating the trade-off.