On an Image (Using SDK)

You can run fine-tuned models on images using Inference.

An Inference server will manage inference. Inference can be run on your local machine, a remote server, or even a Raspberry Pi.

If you need to deploy to the edge, you can use a device like the Jetson Nano. If you need high-performance compute for batch jobs, you can deploy Inference to a server with a GPU.


Follow our Run a Fine-Tuned Model on Images guide to learn how to find a model to run.


If you haven't already, follow our Run Your First Model guide to install and set up Inference.

First, start an Inference server:

inference server start

Next, create a new Python file and add the following code:

from inference.models.utils import get_roboflow_model
from PIL import Image

model = get_roboflow_model(

result = model.infer("path/to/image.jpg")


Replace your API key, model ID, and model version as appropriate.

Then, run the code. You will see predictions printed to the console in the following format:

    "predictions": [
            "class": "rock",
            "confidence": 0.9999997615814209,
            "height": 0.9999997615814209,
            "width": 0.9999997615814209,
            "x": 0.0,
            "y": 0.0

You can plot predictions using supervision. You can install supervision using pip install supervision. Add the following code to the script you created to plot predictions from Inference:

import supervision as sv

detections = sv.Detections.from_roboflow(results)
labels = [p["class"] for p in predictions["predictions"]]

box_annotator = sv.BoxAnnotator()

annotated_frame = box_annotator.annotate(

sv.plot_image(image=annotated_frame, size=(16, 16))