Find a Model on Universe

You can find pre-trained models on Universe for use in your computer vision project.

Over 50,000 datasets on Universe also have trained models associated with the dataset. These models can be used in your computer vision project with Workflows.

To find a dataset on Universe, type in what you want to identify into the search bar. On the search page, check the "Has a Model" box:

This will refine your search to only datasets that contain trained models.

Click on a result to find out more about the project.

The dataset home page will show metrics. These metrics will depend on the model type. For object detection, for example, you will see mAP, precision, and recall; for classification you will see accuracy.

You can upload an image into the "Try This Model" section to try the most recent model version.

You can also click "Models" in the sidebar to see all trained models and choose one to try:

When you open the Models page, you will see the results from an example image in the project test set plotted onto the image:

The class "occupied forklift" is identified in the image.

Deploy a Universe Model

Each Universe model page has a "Deploy Model" button that opens a modal with several deployment options:

• Code Snippet: ready-to-use code in Python, JavaScript, or cURL for calling the model via the Serverless Cloud API. Each snippet includes a "Copy prompt" button that copies a structured prompt you can paste into an AI coding agent.

• MCP: setup instructions for connecting the Roboflow MCP Server to Claude Code, Cursor, or Codex. Once connected, your agent can run inference on Roboflow models directly.

• Custom Workflow: pre-built Workflow templates for common tasks like detection, counting, or visualization.

You can also type a prompt directly in the Roboflow Agent input at the top of the modal to start a guided conversation about building with the model.

Use a Universe Model in a Workflow

You can use Universe models in Workflows.

To use a model, open Roboflow Workflows, create a Workflow, then add a model block:

Click "Model" in the block editor to open the model selector:

Copy the model ID from the top of the Universe model page, then type or paste it into the model selector's search box. Model IDs have the form workspace/project or workspace/project/version.

Roboflow checks the ID against your workspace and Universe, then fills in the model's name, classes, task type, and sample images. Select the matching result to use it. If the ID does not resolve to a real model, a "not found" message shows and the model cannot be selected.

The same model selector is used in the deploy flow, so you can enter a custom model ID anywhere you pick a model.

Run a Universe Model in Code

A Universe model ID works anywhere your own model IDs work. Copy the ID from the top of the model page, then pass it as model_id:

pip install inference-sdk supervision opencv-python

import os
import cv2
import supervision as sv
from inference_sdk import InferenceHTTPClient

image = cv2.imread("forklift.jpg")

client = InferenceHTTPClient(
    api_url="https://serverless.roboflow.com",
    api_key=os.environ["ROBOFLOW_API_KEY"],
)

results = client.infer(image, model_id="forkliftproject/2")

detections = sv.Detections.from_inference(results)
annotated = sv.BoxAnnotator().annotate(scene=image, detections=detections)
annotated = sv.LabelAnnotator().annotate(scene=annotated, detections=detections)
sv.plot_image(annotated)

Set api_url to https://serverless.roboflow.com for the Serverless Cloud API, http://localhost:9001 for a local Inference server, or your Dedicated Deployment URL.

To run a model without picking one from Universe, see Pretrained Model Aliases.