A Clarifai alternative for computer vision
Clarifai’s API went offline in 2026, after Nebius hired its team. Pictograph finds any object you can name with SAM3, trains custom models on your images, and serves them as GPU endpoints for images and live video.
Pictograph vs Clarifai
What a Clarifai computer vision integration did, and what replaces it on Pictograph.
| Feature | Pictograph | Clarifai |
|---|---|---|
| API available to new projects | No: offline since 2026 | |
| Find objects with no training | SAM3 text, box, or point prompt | Pretrained models |
| Train custom models | RF-DETR, YOLOX, Segformer, classifiers | |
| Label images with AI assistance | ||
| Deploy a model as an API | TensorRT on T4, L4, A100 | |
| Ready-made moderation and face models | – |
Where Pictograph fits
Applications that used Clarifai to find objects in images or video, or to run a model trained on their own data. SAM3 covers detection with no training, and a model you train runs as a TensorRT endpoint on a GPU in us-east or us-west, taking single images or live RTSP and WebRTC video.
Where Clarifai led
A broad catalog of ready-made models, including content moderation and face recognition, which Pictograph does not offer, and in its later years inference for large language models as well. If one of those was the core of your integration, look for a provider that specializes in it.
Find objects now, bring your data over
SAM3 answers a text prompt on any image; a COCO export of your Clarifai data imports in two calls.
# Any class you can name, in one image - no training, nothing saved
pictograph sam3 predict shelf.jpg --text "price tag" --json
# A COCO export of a Clarifai dataset: upload the images, then the labels
pictograph images upload-directory shelf-products ./images
pictograph annotations import-coco shelf-products ./instances.jsonfrom pictograph import Client
client = Client()
# Any class you can name, in one image: SAM3 needs no training and saves nothing.
result = client.sam3.predict(
image="shelf.jpg",
text="price tag",
)
for annotation in result.annotations:
print(annotation.name, annotation.type, annotation.confidence)
# A COCO export of a Clarifai dataset: upload the images, then the labels.
client.images.upload_from_directory(
dataset_name="shelf-products",
directory="./images",
)
report = client.annotations.import_coco(
dataset_name="shelf-products",
coco="./instances.json",
)
print(report.images_matched, "matched;", report.unmatched_files, "unmatched")The SAM3 API takes a file, a URL, or bytes and a text, box, or point prompt. The annotations reference also imports YOLO and Pascal VOC labels.
Move off the Clarifai API in four steps
Start with SAM3 for detection today, then replace it with a model trained on your own images.
-
Replace pretrained detection
Call SAM3 with a short phrase ("price tag", "forklift") on any image. It returns a mask or box for every match, with no training and nothing saved.
-
Bring your labeled data
Upload your images and import a COCO export of your annotations, or draft new labels with SAM3 and review them.
-
Train a model
Export the reviewed dataset and train an RF-DETR, YOLOX, or classification model on it on a managed GPU.
-
Deploy it
Serve the model as a GPU endpoint for single images or live RTSP and WebRTC video, or run it on your own hardware.
The training guide walks through a run, and deployments and streams cover serving it.
Clarifai alternative FAQ
What happened to Clarifai?
In May 2026 Nebius announced that Clarifai’s core engineering team, including its founder and CEO, was joining Nebius, and that it was licensing Clarifai’s inference and compute orchestration technology for its own inference service. As of October 2026, Clarifai’s API and documentation sites no longer respond. Applications that called the Clarifai API need another provider.
Can Pictograph detect objects without training a model, like Clarifai’s pretrained models?
Yes, with SAM3. Send any image with a short noun phrase and SAM3 returns a mask or box for every match, named after the phrase; it needs no training and saves nothing. It costs about a cent per image. For a fixed set of classes at video speed, train a small detector instead: it is faster and cheaper per frame once trained.
How do I move my Clarifai datasets to Pictograph?
If you exported your datasets in COCO format, upload the images to a Pictograph dataset and import the COCO file: annotations are matched to images by filename and missing classes are created. If you only have images, SAM3 can draft the labels from text prompts for you to review. Then export the dataset and train on the export.
Does Pictograph do content moderation or face recognition?
No. Pictograph has no ready-made moderation, face recognition, or celebrity models. It is built for detecting, segmenting, and tracking the objects your application cares about, with models you train on your own images. If moderation or face identification is the requirement, choose a provider that specializes in it.
Can Pictograph run on video?
Yes. A deployment can pull an RTSP camera or take WebRTC video from a browser, phone, or app, and returns every frame’s detections as it is processed. Workflows add object tracking, line-crossing counts, dwell time, and webhooks.
Other comparisons
Replace the Clarifai API
$5 of free compute every month. No card required.