AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified Apache-2.0 Self-run

Remove Bg

skill-zju-real-easel-remove-bg · by ZJU-REAL

图片去背景 / 抠图 / 换背景:用 AI 语义分割把主体从背景抠出,输出透明 PNG,或直接换成纯色(电商白底)/ 新场景背景。无需绿幕。当用户说 去背景、抠图、抠图换背景、去掉背景、透明背景、白底图、换背景、抠人像、抠产品、扣图、KO图、商品白底 时使用。基于 shared/scripts/remove_bg.py(rembg u2net)。与 green-screen 区别:green-screen 处理绿幕视频,本 SKILL 处理任意图片;与 ecom-details-image 区别:那个出电商视觉方案,本 SKILL 只做抠图。

No reviews yet
0 installs
1 views
0.0% view→install

Install

$ agentstack add skill-zju-real-easel-remove-bg

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-zju-real-easel-remove-bg)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
2d ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Remove Bg? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

图片去背景 / 抠图换背景

> 用 rembg(AI 语义分割)把主体抠出,输出透明 PNG 或换新背景。无需绿幕。全部走 > skills/shared/scripts/remove_bg.py

> 绿幕视频抠像见 green-screen;电商视觉方案见 ecom-details-image; > 常规缩放/裁切/水印见 image-editing

前置

首次运行自动下载模型(~4-170MB,视模型而定),需外网代理(脚本读 EASEL_PROXY/http(s)_proxy,未设则直连)。先自检:

python skills/shared/scripts/remove_bg.py check

输入

| 字段 | 必填 | 说明 | |------|------|------| | 图片 | 是 | 要抠图的图片(没给就问) | | 输出背景 | 否 | 透明(默认)/ 纯色 / 换图片背景 | | 模型 | 否 | 通用 / 人像 / 精细,见下表 |

输出(outputs/主题名/

  • 抠好的图片(透明用 .png
  • 报告:所用模型、背景形态

执行步骤

脚本路径(相对项目根):skills/shared/scripts/remove_bg.pyremove -h 看参数)。

# 透明背景(务必输出 .png)
python skills/shared/scripts/remove_bg.py remove -i product.jpg \
  -o outputs/主题名/cutout.png

# 电商白底主图
python skills/shared/scripts/remove_bg.py remove -i product.jpg \
  -o outputs/主题名/white.jpg --bg-color white

# 换新场景背景
python skills/shared/scripts/remove_bg.py remove -i person.jpg \
  -o outputs/主题名/scene.png --bg-image scene.jpg

# 人像 + 毛发边缘细腻
python skills/shared/scripts/remove_bg.py remove -i portrait.jpg \
  -o outputs/主题名/cut.png --model u2net_human_seg --alpha-matting

模型选择

| 模型 | 适用 | |------|------| | u2net(默认) | 通用主体 | | u2netp | 轻量快速(质量略低) | | u2net_human_seg | 人像专用 | | isnet-general-use | 更精细的通用分割 | | silueta | 体积小的通用模型 |

抠不干净/边缘毛糙时:换更精细的模型,或加 --alpha-matting(慢但边缘更好,适合头发/毛绒)。

规则

  1. 要透明背景必须输出 .png(jpg 不支持透明,脚本会自动改白底并提示)。
  2. 电商主图用 --bg-color white;换场景用 --bg-image(自动等比覆盖裁切)。
  3. 人像优先 u2net_human_seg,商品/通用用 u2net
  4. 抠图质量取决于主体与背景对比度;复杂/低对比图无法保证完美。
  5. 产物统一进 outputs/主题名/

参考来源

去背景用 rembg(u2net 系列显著性目标检测/分割模型)+ onnxruntime CPU 推理,是无 GPU 抠图的 主流方案。换背景合成用 Pillow alpha_composite。把模型加载、代理注入、背景合成封装成确定性脚本。

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.