SAM3 LiteText

SAM3 LiteText replaces SAM3's large text encoder with a compact MobileCLIP student, so concept prompts like “apple” or “forklift” run comfortably on laptops. You get the full SAM3 prompt toolkit for masks, boxes and rotated boxes at a fraction of the memory.

Object DetectionInstance SegmentationSemantic SegmentationOriented Object DetectionEfficientSAM3Custom license
Recommended10/10
VRAM2.2–2.5 GBweights, plus ~0.5 GB reserve
Tools6across 4 annotation types
Runs onmacOSCUDACPU4 of 4 runtime profiles

Install

boat model install sam3-litetext

Without a tag BOAT installs sam3-litetext:S0. Tags install side by side, so you can keep several.Download: 2.2 GB of weights.

You will be asked to accept the model's license on first install. Add --accept-license to accept it non-interactively. Read the terms

Tools

What SAM3 LiteText adds to the annotation editor. Each example loops what you do and what comes back.

Text to Mask example on instance segmentation

Text to Mask

Text promptMasks

Type a class name such as “apple” and press Enter. BOAT previews every matching mask before you apply them.

  • Preview, then apply
Text to Boxes example on object detection

Text to Boxes

Text promptBounding boxes

Object Detection

Type a class name such as “car” and press Enter. BOAT previews every matching bounding box before you apply them.

  • Preview, then apply
Box prompt example on instance segmentation

Box prompt

BoxMasks

Draw one or more boxes — add negative boxes to exclude areas — to segment exactly those objects.

  • Preview, then apply
  • Multiple prompts
  • Negative prompts
Point prompt example on instance segmentation

Point prompt

PointsMasks

Click on the object, and add negative points to carve out what doesn't belong, to get a precise mask.

  • Preview, then apply
  • Multiple prompts
  • Negative prompts
Auto search example on instance segmentation

Auto search

BoxMasks

Draw a loose box around an object. The model finds the object inside and returns a tight mask.

  • Preview, then apply
Auto search example on oriented object detection

Auto search

BoxRotated boxes

Oriented Object Detection

Draw a loose box around an object. The model finds the object inside and returns a tight rotated box.

  • Preview, then apply

Tags

Each tag is an installable variant, such as a size or quantization, with its own weights.

TagWeightsPlatformsInstall
sam3-litetext:S0default2.2 GB · 1 filemacOSCUDACPU
sam3-litetext:S12.3 GB · 1 filemacOSCUDACPU
sam3-litetext:L2.5 GB · 1 filemacOSCUDACPU

Platforms

BOAT picks the runtime profile that matches your machine. Only that profile's code is downloaded.

macOS · Metal

Apple silicon (MPS)

Python 3.11 · 1.8 MB

Linux · CUDA 12.8

NVIDIA, incl. Blackwell

Python 3.11 · 1.8 MB

Linux · CUDA 12.6

NVIDIA GPUs

Python 3.11 · 1.8 MB

Linux · CPU

x86_64 CPU

Python 3.11 · 1.8 MB

License

BOAT's adapter code is Apache-2.0. The model's own code and weights keep their upstream license:Custom.

License terms shown at install

SAM3 is provided under Meta’s SAM License.

By continuing, you acknowledge that your use of SAM3 must comply with all applicable laws, regulations, and Trade Controls. You confirm that you are not the target of Trade Controls and that you will not use, or permit others to use, SAM3 for:

- activities subject to the International Traffic in Arms Regulations (ITAR);

- military or warfare purposes;

- nuclear industries or applications;

- espionage; or

- the development or use of guns or illegal weapons.

Bundled license files: BOAT-APACHE-2.0.txt, EFFICIENTSAM3-LICENSE.txt, NOTICE.md, UPSTREAM-LICENSE.txt

Package details

Package
sam3-litetext
Version
0.0.1
Type
Common model
BOAT Core
≥ 0.1.0 · < 1.0.0
Idle unload
300s (change with boat model setup)
Inference timeout
600s
Manifest SHA-256
d7866026581ccc061cb2e290ab7a2db7572d6bc11c2393e1af1b61de5a9e1e79
Upstream notice
BOAT adapter code is Apache-2.0. Upstream model code and weights keep their own licenses.

Model: sam3-litetext (SAM3-LiteText, MobileCLIP-S0 ctx16 student text encoder)
Pinned weight provenance is in ../weights.json. Vendored code revisions, when present,
are recorded in ../vendor/UPSTREAM_REVISION. The local macOS adaptations are retained.
Runtime dependencies are installed from pinned distributions with their own license files.
↑↓ navigate↵ open