sam3-segmentation skill
sam3-segmentation outlines photo objects with SAM 3 and saves masks, raw candidates and review decisions for an agent workflow.
- 128 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Following the skill, the agent runs a local model, downloads its files and saves masks; sending photos to external reviewers requires separate user approval.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/sam3-segmentationDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Following the skill, the agent runs a local model, downloads its files and saves masks; sending photos to external reviewers requires separate user approval.
This is third-party code. Review the repository files before installing.
What it does
sam3-segmentation gives agents instructions and commands for outlining photo objects with SAM 3. Select objects by text, click or box, or find similar instances, then save masks for review and export. A detailed workflow for small objects uses overlapping crops, retained raw candidates, review sheets and reversible decisions. A viewer supports boundary inspection. Masks remain model proposals; object counts require human verification before they can be treated as confirmed.
Who it is for. Image researchers outlining photo objects and retaining inspectable intermediate results.
Good fit when
- Outline an object from a text prompt or box.
- Review many small objects through crops and saved candidates.
- Retain reversible mask-removal decisions.
- Export reviewed masks as PNG or COCO RLE with a viewer.
Not a fit when
- Segmentation must run on CPU or Apple GPUs only.
- You need video or medical-image processing.
- Model masks would be treated as verified ground truth.
Example request
Outline seeds in this photo, retain every candidate and a review sheet, and list uncertain masks before counting.Limitations
Segmentation requires an NVIDIA GPU on Linux or WSL2; the no-GPU demo only displays generated shapes. Accepted SAM 3 access on Hugging Face and disk space for the model are required, with the owner logging in separately. The author does not claim thorough human validation, and predictions can confuse background, parts and whole objects. Accurate counts need human-reviewed masks and a defined review-completion criterion that the current workflow does not guarantee.
How to disable. Stop model and viewer processes, then remove the skill. Review environments and outputs outside the skill folder separately.
FAQ
Does segmentation work on a Mac?
Not on CPU or Apple GPU. Without NVIDIA, use the generated-shape demo to inspect formats.
Should I give the agent my token?
No. Log in yourself with `hf auth login`; do not paste a token into chat.
Can I undo mask removal?
Yes, when the agent retains review decisions as instructed. Inspect rejected-mask records before export.
What does the no-GPU demo show?
Generated shapes, the viewer and export; it does not segment your photos.
Does mask count equal the true object count?
No. Manually check misses, duplicates and mistaken object parts.
Discussion
No questions or reviews yet.
Related
SkillsThe scan-to-threejs skill refines a 3D scene in a loop, checking renders against room photos from matching camera angles.
rea-video is an agent skill that breaks a video down shot by shot, calling out real versus AI footage and building a remake plan.
A local MCP server for editing and analyzing images: the agent crops frames, finds objects and reads text from photos
Six skills that let the agent turn video into a short Chinese narration recap, locally via ffmpeg
A skill rebuilds the object in a reference image as a procedural Three.js model written in code, not downloaded art
Two skills for generative video and images: video writes Seedance, Kling and Veo prompts by dramaturgy rules, image picks the model and its syntax
README in short
Environments live outside the skill and survive updates. SAM 3 and photo rights are separate from the skill license.
README badge
Original SKILL.md
---
name: sam3-segmentation
description: >-
Outline, mask, cut out or count objects in images with Meta's SAM 3 (Segment
Anything Model 3), run locally on an NVIDIA GPU. Use when the user wants pixel
masks or outlines of things in a photo, even without saying "segmentation" or
"SAM": every instance of a phrase ("all the dogs"), one object from a click or
box, automatic "segment everything", or careful, auditable annotation of many
small objects or parts, such as corn kernels, seeds, cells, insects or leaves,
with COCO RLE / PNG masks, review and a zoomable viewer. Prefer it to judging
outlines, counts or object positions by eye. Includes first-time setup and a
demo that needs no GPU. Outputs are model proposals, not verified
ground truth. Not for classifying, captioning or editing images, video,
medical scans, or customer and network segmentation.
license: MIT
compatibility: >-
Model inference needs Linux (or Windows WSL2), an NVIDIA GPU with drivers for
CUDA 12.8 (12-24 GB memory suggested), uv, Python 3.12, about 11 GB of
downloads plus room for run records, and the user's own approved Hugging Face
access to facebook/sam3. No CPU or Apple
GPU inference. The demo, review and export tools run anywhere uv runs.
metadata:
version: "0.1.0"
author: Aurélie Guisnet
repository: https://github.com/lillyguisnet/sam3-segmentation
---
# SAM 3 segmentation
## Choose a mode
**Direct SAM calls** (`scripts/sam.py`): the model and nothing else. Use them
for a quick outline, a few objects, a "segment everything" pass, exploring a
photo before a full run, or your own procedure. Read