- Category
- AI Tools
- Rank
- No. 1801Tools index
Previous survey · No. 1808 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- elebumm
- GitHub
- 3 stars
- Date
About
Text-prompted video segmentation CLI using SAM 3.1 via fal.ai, Replicate, or local GPU, exporting ProRes 4444 RGBA.
What it does
Samify turns a named subject in footage into a transparent layer for compositing. It asks one of three segmentation services to build a black-and-white mask for every frame, then combines that mask with the source video. Users can preview results, refine mask edges, preserve audio, cache masks, and choose editing or web-oriented output.
Why it's ranked here
The case for Samify is practical rather than broad. It joins subject detection, mask refinement, quality reporting, caching, and alpha encoding in one command-line workflow. Cloud and offline choices reduce deployment lock-in. However, backend capabilities differ, and the default service ignores negative prompts, so identical commands do not always mean identical behavior.
What's good
The shared mask pipeline keeps encoding behavior consistent across three backends. Preview and cost estimation reduce wasted processing, while cached masks let users change feathering, morphology, or output format without repeating segmentation. Detection coverage and gap timestamps expose weak tracking. Batch processing, preserved source audio, and three alpha-capable output choices fit real post-production work.
Tradeoffs
Cloud use requires an API key, internet access, uploads, and usage fees. Offline processing adds large model weights, local dependencies, and substantial GPU memory pressure. The subject should appear in the first frame for reliable tracking. Replicate has an approximate 256 MB inline upload limit, while the default backend ignores negative prompts. Text overlays can also fragment masks.
How to use it well
Use Samify for editors, motion designers, and technical creators who repeatedly isolate people or named objects from short clips. Start with a noun prompt and preview the mask. Cache accepted masks before tuning edges or formats, and inspect reported tracking gaps. Choose cloud processing for convenience or long footage, and local processing for offline work. It does not cover general scene segmentation without a specific subject.
Technical notes+
pyproject.toml defines a setuptools package for Python 3.10 or newer, a samify console entry point, core cloud dependencies, and separate fal and local extras. samify/cli.py orchestrates backend selection, probing, caching, previews, batch handling, refinement, and output selection. samify/video_io.py shells out to ffmpeg and ffprobe, aligns source and mask timestamps, scales masks to the source, applies morphology and blur, then performs alpha merging. samify/backends/__init__.py provides the shared result contract and environment-based backend resolution. samify/backends/fal_cloud.py decodes per-frame RLE, samify/backends/replicate_cloud.py downloads a ready mask video, and samify/backends/local.py loads frames and propagates masks with optional VRAM-based chunking. One detail deserves a stronger integration fixture: samify/backends/fal_cloud.py describes column-major RLE but reshapes the decoded buffer using the default array order, while tests/test_unit.py checks flattened offsets rather than spatial orientation. tests/test_smoke.py covers real Replicate and local paths conditionally.
Observed
- License
- Samify is MIT-licensed; the SAM models are identified as Apache 2.0.
- Primary language
- Python
- Packaging
- Setuptools package configured through pyproject.toml for Python 3.10 or newer, with fal and local optional dependency groups.
- Interface
- Command-line application exposed through the samify console command and Python module entry point.
- Backends
- Supports fal.ai, Replicate, and local Hugging Face Transformers execution.
- Runtime tools
- Requires ffmpeg and ffprobe for video probing, mask composition, previews, and encoding.
- Output surfaces
- Produces ProRes 4444 MOV, VP9 alpha WebM, or RGBA PNG sequences.
- Test structure
- Repository includes unit tests for core logic and conditional end-to-end smoke tests for Replicate and local execution.
Read from README.md, pyproject.toml, samify/cli.py, samify/__init__.py, samify/__main__.py, samify/video_io.py, samify/backends/local.py, samify/backends/__init__.py, samify/backends/fal_cloud.py, samify/backends/replicate_cloud.py, tests/test_unit.py, tests/test_smoke.py, .claude/settings.local.json, .claude/skills/samify/SKILL.md.
What it can do
Segment video using text prompts
Video file and text description → Segmented video regions
Process video segmentation via fal.ai API
Video file and text prompts → SAM 3.1 segmentation results
Process video segmentation via Replicate API
Video file and text prompts → SAM 3.1 segmentation results
Process video segmentation on local GPU
Video file and text prompts → SAM 3.1 segmentation results
Export segmented video to ProRes 4444 RGBA format
Segmentation results → ProRes 4444 RGBA video file
Run video segmentation from command line interface
CLI commands with video and prompt parameters → Processed segmented video
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.
