Data labeling for modern AI teams.

Distill frontier intelligence into custom vision models.

200 free AI credits / month. No payment method required.

pipeline.py

from pictograph import Client
from pictograph.workflows import full_pipeline

client = Client()

# Upload images, annotate with SAM3, train an RF-DETR model
report = full_pipeline(
    client,
    dataset_name="Doorbell Camera",
    folder="./doorbell_camera_images",
    classes=[("person", "polygon"), ("package", "polygon")],
    pipeline="rfdetr_segmentation",
)

Trained model RF-DETR · 15 epochs · A100 GPU

auto_annotate.py

from pictograph import Client

client = Client()

# Three SAM3 prompt modes — point, box, or text
result = client.auto_annotate.text(
    dataset_name="Road Signs",
    image_filename="img-001.jpg",
    text="stop sign",
    name="stop_sign",
)

# Saved annotation polygon
client.annotations.save(image_id, result.annotations)

AI Annotation

SAM3 auto-annotation

Generate pixel-perfect annotations from point, box, or text prompts.

Point Click to add positive and negative prompts.

Box Select an object, annotate all similar objects.

Text Enter "Stop sign", annotate all objects matching the description.

Batch Run text prompts over massive datasets in a single job.

Read the auto-annotate guide

search.py

from pictograph import Client

client = Client()

# "Find every image like this one"
similar = client.search.by_similarity(
    image_id="img-001",
    threshold=0.6,
    limit=20,
)

# "Find every image containing this content"
tagged = client.search.by_tag(
    objects=["car", "person"],
    scenes=["street"],
)

Visual search

Visual content indexing
and search

Automatic image embedding and content tagging on upload. Search by visual similarity or objects, scenes, styles. Enterprise grade data privacy and security guaranteed.

Automatic Embeddings and tags generated for every image at upload

Self-hosted Embeddings generated on closed Pictograph-managed hardware

By image Millisecond latency visual similarity search

By content Objects, scenes, and attributes auto-detected

Read the search guide

Model training

Train custom vision models,
in one call

Pick a vision task and hardware; The SDK exports your dataset, runs a training job, and returns the trained model. No ML infrastructure to manage.

YOLOX Object detection, trains quickly, edge-deployable

Detectron2 Instance segmentation with pixel masks

Segformer Semantic segmentation, pixel-wise class maps

RF-DETR State of the art performance, detection and instance segmentation

Read the training guide

train.py

from pictograph import Client
from pictograph.workflows import train_pipeline

client = Client()

# Export, train, fetch the model — one call
run, model = train_pipeline(
    client,
    "road-signs",
    pipeline="yolox",   # or detectron2 / rfdetr / …
    gpu="a10g",
)

client.models.download(model.id, "./yolox.onnx")

Everything around the editor

Pictograph is a developer-friendly stack; build your own workflow, or let an agent drive.

Multi-format export

COCO, YOLO, CVAT, Pascal VOC, LabelMe, CSV, or canonical Pictograph JSON.

Team workspaces

Role-based access — owner, admin, member, viewer. Audit log included.

Python SDK + CLI

Typed Pydantic models, auto-retry, idempotency keys. `pip install pictograph`.

Agent-ready

Drive every operation from Claude or OpenAI via the 28-tool registry.

Label faster, start shipping today