AI annotation

Auto-annotate with SAM3: point, box, and text prompts.

Turn a click, a box, or a text phrase into pixel-perfect masks. Label one image, or run a batch over thousands. Every result stays editable.

Start free Read the guide

auto_annotate.py

from pictograph import Client

client = Client()

# Type a phrase, get masks back, zero-shot
result = client.auto_annotate.text(
    image_id="img_market_01",
    prompt="ripe tomato",
)

for ann in result.annotations:
    print(ann.name, ann.type)   # "ripe tomato" polygon
    print(ann.polygon.paths)    # editable polygon points

How it works

From a prompt to editable annotations

No setup and no model to wire up. Pick how you want to describe the object, and SAM3 does the segmentation.

Choose a prompt mode

Click a point, drag a box, or type a text phrase. Each mode targets a different way of describing what to label.
SAM3 returns masks

A managed GPU runs SAM3 and sends back pixel-precise masks as polygons and bounding boxes.
Edit, or run a batch

Refine results in the editor, or run a batch over a whole dataset.
Export or train

Send the labels to a COCO or YOLO export, or train a model on them.

Three prompt modes

Point, box, or text, your choice

Every mode runs the same SAM3 model and returns the same editable polygons and boxes. Mix them on a single image as you go.

Point

point.py

from pictograph import Client

client = Client()

# Positive points include, negative points exclude
result = client.auto_annotate.point(
    image_id="img_market_01",
    points=[
        {"x": 412, "y": 280, "label": 1},
        {"x": 90, "y": 120, "label": 0},
    ],
)

Box

box.py

from pictograph import Client

client = Client()

# Draw one box around an object
result = client.auto_annotate.box(
    image_id="img_market_01",
    box={"x": 360, "y": 220, "w": 120, "h": 140},
)

Text

text.py

from pictograph import Client

client = Client()

# Label by concept, no training required
result = client.auto_annotate.text(
    image_id="img_market_01",
    prompt="bell pepper",
)

What each mode is for

Pick the prompt that fits the job

Single objects, lookalikes, whole concepts, or an entire dataset. Each mode maps to a different labeling task.

Point Click to add and exclude regions.

Box Draw one, label every match.

Text Type a prompt, label by concept.

Batch Label thousands of images in one job.

Manual annotation is always free. Draw bounding boxes, polygons, polylines, and keypoints by hand at no cost, and refine any auto-annotation result the same way.

Keep exploring

Train a model Turn your labeled dataset into a model: YOLOX, RF-DETR, Segformer, and more on managed GPUs. Deploy a model Stand up a trained model as one authenticated /predict endpoint, billed by uptime. Build a workflow Chain source, model, filter, track, and sink nodes over images, video, or a dataset.

FAQ

Auto-annotation FAQ

What is SAM3 auto-annotation?

SAM3 auto-annotation generates polygon or bounding-box labels from a point, box, or text prompt in one API call. Click an object, drag a box, or type a class name, and the model returns precise masks you can save as annotations.

Can I auto-label a whole dataset at once?

Yes. Batch auto-annotation runs SAM3 text prompts across many images and classes in one job, with built-in duplicate detection so overlapping detections of the same class are merged. It bills per image and class from compute credits.

Does auto-annotation support text prompts?

Yes. Type a phrase like "car" or "traffic light" and SAM3 grounds it to every matching region in the image, returning a mask for each. This works on a single image interactively or across a dataset in batch.

How accurate is SAM3 segmentation?

SAM3 produces pixel-accurate masks with hole support, rendered with an even-odd fill rule. You can refine any result by adding positive or negative points, or by editing the polygon vertices directly.

Label your dataset in minutes

$5/mo free compute. No payment method required.

Start free Read the docs