mirror of https://github.com/snapotter-hq/SnapOtter.git synced 2026-08-03 07:46:42 +02:00

Files

T

SnapOtterandGitHub ec98dfde9a feat(landing,docs): SEO fixes and query-matched tool pages (#383 )

Landing:
- Preload the Bricolage display font (LCP heading) to cut first paint
- Cap tool-page meta descriptions at ~160 chars via whole-sentence
  truncation, so SERP snippets no longer overflow or cut mid-phrase
- Drop dead HowTo + FAQPage JSON-LD from tool pages (Google removed
  HowTo rich results in 2023 and FAQ rich results in 2026); the visible
  content stays. Breadcrumb + WebApplication schema retained
- Shorten 22 tool titles that exceeded the ~60-char SERP limit
- Add an optional h1 override to tool SEO data
- Rewrite adjust-colors, smart-crop, and ringtone-maker to match real
  Search Console query intent (saturation; object-aware cropping;
  convert audio to ringtone), with query-matched H1s and FAQs

Docs:
- Add meta descriptions to 20 pages missing frontmatter

Repo:
- Ignore local credential JSON files (.secrets/, *client_secret*, *service_account*)

2026-07-01 16:20:05 +08:00

3.3 KiB

Raw Blame History

description

description
Subject-, face-, and entropy-aware cropping that frames images intelligently using Sharp and AI face detection.

Smart Crop

Smart subject-aware, face-aware, or trim-based cropping. Uses Sharp's attention/entropy strategies and AI face detection for intelligent framing.

API Endpoint

POST /api/v1/tools/image/smart-crop

Processing: Asynchronous (returns 202, poll /api/v1/jobs/{jobId}/progress for status via SSE)

Model bundle: face-detection (200-300 MB) - required only for face mode

Parameters

Parameter	Type	Required	Default	Description
file	file	Yes	-	Image file (multipart)
mode	string	No	`"subject"`	Crop mode: `subject`, `face`, `trim`. (Legacy values `attention` and `content` map to `subject` and `trim`)
strategy	string	No	`"attention"`	Strategy for subject mode: `attention` or `entropy`
width	integer	No	-	Target width in pixels
height	integer	No	-	Target height in pixels
padding	integer	No	`0`	Padding percentage around subject (0-50)
facePreset	string	No	`"head-shoulders"`	Face framing preset: `closeup`, `head-shoulders`, `upper-body`, `half-body`
sensitivity	number	No	`0.5`	Face detection sensitivity (0-1)
threshold	integer	No	`30`	Trim mode threshold for background detection (0-255)
padToSquare	boolean	No	`false`	Pad trimmed result to a square
padColor	string	No	`"#ffffff"`	Background color for padding
targetSize	integer	No	-	Target size for padded output (pixels)
quality	integer	No	-	Output quality (1-100)

Example Request

curl -X POST http://localhost:1349/api/v1/tools/image/smart-crop \
  -F "file=@portrait.jpg" \
  -F 'settings={"mode":"face","width":1080,"height":1080,"facePreset":"head-shoulders"}'

Response

Initial Response (202 Accepted)

{
  "jobId": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "async": true
}

Progress (SSE at `/api/v1/jobs/{jobId}/progress`)

event: progress
data: {"phase":"processing","percent":50}

Final Result (via SSE)

{
  "phase": "complete",
  "percent": 100,
  "result": {
    "jobId": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
    "downloadUrl": "/api/v1/download/{jobId}/portrait_smartcrop.jpg",
    "originalSize": 500000,
    "processedSize": 320000
  }
}

Modes

Subject Mode

Uses Sharp's attention or entropy strategy to find the most visually interesting region and crops around it.

Face Mode

Detects faces using AI, then frames the crop around detected faces using the specified facePreset. Falls back to subject mode (attention strategy) if no faces are detected.

Trim Mode

Removes uniform borders/background from the image. Optionally pads the result to a square with a specified background color and target size.

Notes

This tool uses the createToolRoute factory with executionHint: "long", so it returns 202 with SSE progress.
Face mode requires the face-detection model bundle (200-300 MB).
Subject and trim modes work without any AI model bundle.
The facePreset determines how tightly the crop frames detected faces: closeup is the tightest, half-body is the widest.
If no width/height are specified, defaults to 1080x1080.

3.3 KiB Raw Blame History