Skip to main content
Anastig · visual AI · early access

Build visual AI from your own images.

Annotate and export model-ready datasets in Anastig Studio, then run detection, segmentation, and prompt models privately in Vision Lab — on a worker you control.

Private by default · runs on a worker you control · not used to train shared models

Two products, one pipeline

Annotate in Studio. Run models in Vision Lab.

One pipeline: build datasets with your own labels, privately — then run and review models on them. Switch between the two below.

anastig-studio · annotate
scratch · 0.97
dent · 0.91
metal-panel-01 · 800×4502 objects · box
Dataset-to-model workspace

Anastig Studio

Upload your images, draw and refine annotations with AI assist, and export a clean, validated dataset — with your own labels, on your terms.

  • Annotate with your own classes — boxes, polygons, keyboard-fast.
  • AI-assisted: SAM refine, one-example matching, honest review queues.
  • Export YOLO, COCO, and Pascal VOC + dataset.yaml — validated before download.
How the two connect

From raw images to reviewed results.

  1. 01
    Studio

    Upload images

    Drop in photos or use samples. Originals stay in your browser in demo mode.

  2. 02
    Studio

    Annotate & refine

    Draw boxes and polygons, or let SAM and one-example matching propose them.

  3. 03
    Studio

    Export dataset

    YOLO, COCO, VOC + dataset.yaml — validated before download.

  4. 04
    Vision Lab

    Run models privately

    Detection, segmentation, and prompt models on a worker you control.

  5. 05
    Vision Lab

    Review results

    Compare models, then turn outputs into reviewable annotations.

Why anastig

Built for the team, not the demo.

Your own labels

Define, rename, reorder, and recolor classes that match your real inspection workflow — not someone else's demo dataset.

Private by default

Images stay in your browser in demo mode. In production, your data stays in your tenant and is never used to train shared models. Vision Lab inference runs on a worker you control — your media is not uploaded to Anastig.

Correct exports

YOLO labels, COCO JSON, Pascal VOC, and dataset.yaml. Exports are validated before download — no silent bad data.

Your data stays private by default. We never train shared models on customer datasets without explicit, contract-bound consent. Read the security overview

Start with one inspection task.

Open the demo to annotate with your own images, or request a pilot and we'll scope a complete dataset-to-model workflow with you.