segImage API & SDK

REST API for background removal and sticker cutout. Base URL: . No API key required. Uploads are processed in memory and never stored. Interactive docs: Swagger UI · raw spec: openapi.json.

抠图/去背景 REST API。Base URL：。无需 API Key。上传内容只在内存中处理，从不保存。交互文档：Swagger UI · 规范：openapi.json。

⚠️ This API is for testing only. For commercial use or partnership inquiries, contact stormnetwork@163.com (海口龙华区妙妙鱼传媒工作室 / Miaomiaoyu Media Studio, Longhua District, Haikou). Public instance limits per IP: 1 call / 10 s and 5 calls / 60 s (HTTP 429 when exceeded).

Endpoints

Route	Description
`GET /api/v1/health`	Version, loaded models, limits
`GET /api/v1/models`	rembg models + currently available free vision models
`POST /api/v1/segment`	One subject → transparent PNG (method = rembg \| grabcut \| ai)
`POST /api/v1/multi`	Up to 6 subjects → one PNG each
`POST /api/segment`, `GET /api/health`	Legacy aliases (v1.0/1.1 response shape), kept for compatibility

Form fields (multipart/form-data)

Field	Default	Description
`file`	—	JPEG / PNG / WebP, ≤ 12 MB
`method`	rembg	rembg · grabcut · ai
`rembg_model`	u2netp	u2netp · isnet-general-use · u2net · silueta
`complete`	true	Smart subject completion (pure vision)
`smooth`	3	0–10
`despeckle`	true	Remove small islands, fill holes
`white_border`	true	White sticker outline
`points_json`	—	grabcut: `[{"x":0.5,"y":0.5,"label":1}]` (0..1)
`rect_json`	—	grabcut: `[x,y,w,h]` (0..1)

Response & errors

{ "ok": true, "version": "1.2.1", "method": "rembg", "engine": "rembg_u2netp+smart_complete",
  "rembg_model": "u2netp", "width": 960, "height": 1280, "elapsed_ms": 1800,
  "image_png_b64": "iVBOR…", "mask_png_b64": "iVBOR…", "detection": null }

{ "ok": false, "error": { "code": "FILE_TOO_LARGE", "message": "File larger than 12 MB" } }

HTTP	code
400	EMPTY_FILE · BAD_JSON · UNKNOWN_MODEL
413	FILE_TOO_LARGE
415	UNSUPPORTED_IMAGE
422	VALIDATION_ERROR
500	SEGMENTATION_FAILED

Limits: 12 MB per upload; images are downscaled to ≤1280 px for rembg; the public demo runs on a small 4-vCPU box with no GPU — please keep it to a few requests per minute, and self-host for batch jobs. The ai/multi methods only use free OpenRouter models and may be slow (up to ~2 min) or rate-limited, falling back to rembg.

限制：单次上传 ≤ 12 MB；rembg 会把图片缩到长边 ≤1280px；公共演示跑在一台 4 核、无 GPU 的小服务器上，请控制在每分钟几次，批量任务请自行部署。ai/multi 仅使用 OpenRouter 免费模型，可能较慢（最长约 2 分钟）或被限流，失败时回退 rembg。

curl

BASE=
curl -s $BASE/api/v1/health
curl -s -F file=@photo.jpg -F method=rembg $BASE/api/v1/segment \
  | python3 -c "import sys,json,base64;d=json.load(sys.stdin);open('sticker.png','wb').write(base64.b64decode(d['image_png_b64']))"
curl -s -F file=@photo.jpg -F method=grabcut -F 'points_json=[{"x":0.5,"y":0.5,"label":1}]' $BASE/api/v1/segment
curl -s -F file=@photo.jpg $BASE/api/v1/multi

Python SDK

Single file, standard library only: sdk/python/segimage.py

单文件、仅依赖标准库：sdk/python/segimage.py

from segimage import SegImage, SegImageError

client = SegImage("")
res = client.segment("photo.jpg", method="rembg", smooth=3)
res.save("sticker.png")
print(res.engine, res.elapsed_ms)

for i, s in enumerate(client.multi("desk.jpg").subjects, 1):
    s.save(f"{i}_{s.label}.png")

JavaScript / TypeScript SDK

ESM, uses fetch + FormData; browser and Node ≥ 18; typings included: sdk/js

ESM，基于 fetch + FormData；浏览器与 Node ≥ 18 通用，自带类型：sdk/js

import { SegImage } from "./segimage.js";

const client = new SegImage("");
const res = await client.segment(fileInput.files[0], { method: "rembg" });
document.querySelector("img").src = URL.createObjectURL(res.image);