技能 object-separation

📦

object-separation

Name: object-separation
Author: barty-bart

內容修訂 r1 中風險 📁 檔案系統存取⚡ 包含腳本⚙️ 外部命令

Separate Video Subjects Locally with SAM 2.1

Manual video subject masking can take substantial editing time. This skill guides local SAM 2.1 processing to generate frame masks and a green highlight preview.

支援: Claude Codex Code(CC)

📊 69 尚可

Maintain this Skill? Claim it 📊 69 尚可

使用我的智慧代理安裝

將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。

智能體請求

Review the Skillstore skill "object-separation" from https://skillstore.io/skills/barty-bart-object-separation.md and its manifest at https://skillstore.io/api/skills/barty-bart-object-separation/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.

你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。

Agent 可讀資源

當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。

Markdown 詳情 GET /skills/barty-bart-object-separation.md 已簽名 manifest GET /api/skills/barty-bart-object-separation/manifest 已簽名 lockfile GET /api/skills/barty-bart-object-separation/lockfile

測試它

正在使用「object-separation」。 Separate the presenter in clip.mp4, frames 0 through 19.

預期結果:

Expected preview: work/separated.mp4, showing the presenter highlighted in green.
Expected masks: numbered PNG files in work/masks/subject/, with white pixels marking the subject.
Review the beginning, middle, and end before extending the selected range.

正在使用「object-separation」。 Prepare a product mask without including the hand holding it.

預期結果:

Point plan: positive points inside the product and negative points on the hand and table.
Validation: inspect contact edges and occlusions in a short test range.
Expected deliverables: binary subject masks and a green highlight video; background replacement requires another editing step.

安全審計

中風險

v1 • 2026/10/20 開啟版本化報告

All 27 static alerts are false positives: Markdown formatting, model inference setup, and ordinary local processing do not demonstrate the alleged threats. A separate cleanup instruction could delete unrelated cached models on Windows. No evidence found of prompt injection or video exfiltration; setup downloads dependencies and model files.

已掃描檔案

203

分析行數

複核項目

已忽略的誤報

已確認的安全問題 (1)

中

SKILL.md:157-161

Overbroad Windows Model Cache Cleanup

Optional cleanup says to delete the model but gives the Windows path C:\Users\<name>\.cache\huggingface\hub. Deleting this shared directory could remove unrelated models.

The Windows alternative names the whole hub directory, unlike the model-specific Unix path. The deletion instruction therefore exposes unrelated cached models to removal.

風險因素

📁 檔案系統存取 (5)

README.md:3 SKILL.md:161 SKILL.md:161 SKILL.md:76 SKILL.md:120

⚡ 包含腳本 (1)

SKILL.md:57

⚙️ 外部命令 (20)

審計單位: codex

分享與引用此報告

分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。

開啟版本化報告

複製報告連結

https://skillstore.io/skills/barty-bart-object-separation/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/barty-bart-object-separation/security.svg)](https://skillstore.io/skills/barty-bart-object-separation?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/barty-bart-object-separation?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/barty-bart-object-separation/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/barty-bart-object-separation.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

學術引用 (APA · BibTeX · CFF)

APA 引用

barty-bart. (2026). object-separation security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/barty-bart-object-separation/audits/1

BibTeX 引用

@techreport{barty-bart-barty-bart-object-separation-2026,
  author = {barty-bart},
  title = {object-separation security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/barty-bart-object-separation/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "object-separation security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "barty-bart"
date-released: "2026-10-20"
url: "https://skillstore.io/skills/barty-bart-object-separation/audits/1"
identifiers:
  - type: other
    value: "skillstore:barty-bart-object-separation:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 中

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Prepare Masks for Video Compositing

Generate subject masks from a short clip for later compositing in an external editor.

Isolate Products from Handling

Use exclusion points to keep hands, packaging, or a tabletop outside a product mask.

Inspect Local Segmentation Results

Create binary frame masks and inspect green overlays before reusing the masks in a separate analysis workflow.

試試這些提示

Test a Short Subject Mask

Use object-separation on clip.mp4. Separate the person in frames 0 through 19 and show the green preview before processing more frames.

Separate a Product

Separate the bottle in product.mp4, frames 30 through 89. Exclude the holding hand and table, then save the masks and preview.

Plan for Limited Hardware

Inspect available hardware for interview.mp4 and report the selected model, download size, and expected speed. Wait for approval before installation or downloads.

Refine and Export a Frame Range

For demo.mp4 frames 100 through 219, test 20 frames first. Refine exclusion points around occlusions, then export masks and the green preview. Preserve existing outputs.

最佳實務

Test about 20 frames and inspect occlusions before processing the full selected range.
Confirm download sizes and use a dedicated project directory; preserve existing outputs before rerunning.
Delete only confirmed skill-specific cache folders during cleanup, especially on Windows.

避免

Expecting the supplied renderer to produce transparent video or replace the background automatically.
Processing long CPU-only clips without estimating time or checking a short sample.
Deleting the entire shared Hugging Face cache to remove one downloaded model.

常見問題

Do I need a GPU?

No. The workflow selects a small model for CPU processing, but processing can take about five seconds per frame.

Does the workflow upload my video?

The provided code processes video locally and contains no video upload operation. Initial setup downloads packages and model files.

Which files does it produce?

It writes work/separated.mp4 with a green subject highlight and numbered binary masks under work/masks/subject/.

Does it remove or replace the background?

Not in the provided renderer. It generates masks and a green preview; transparent export or background replacement needs an additional workflow.

How does it handle hair and attached objects?

Masks are hard-edged and can miss fine hair. Add exclusion points for packaging or other attached objects, then review the result.

What should I check before cleanup?

Keep needed outputs and confirm exact cache folders. The Windows cleanup path names a shared cache and could remove unrelated models.

開發者詳細資訊

作者

barty-bart

授權

MIT

Skillstore 修訂

版本提示

作者未聲明版本。

儲存庫

https://github.com/barty-bart/motion-graphics/tree/3d83d987d658601d8a1269eb691622bdcd427335/skills/object-separation

引用

f67574df28ce4d0d72f3d2a644e0e75b89127ef9

維護新鮮度

2026/10/20

使用量

0 次下載 · 0 次瀏覽

檔案結構

📄 README.md

📄 SKILL.md