Vision Detector

There's no need to build an application each time. 

With Vision Detector, running CoreML models on your dievice is effortless.

Vision large language model is also supported in ver.3 via API.

Vision Detector performs image processing using a CoreML model on iPhones and iPads. Typically, CoreML models must be previewed in Xcode, or an app must be built with Xcode to run on an iPhone. However, Vision Detector allows you to easily run CoreML models on your iPhone.

The supported types of machine learning models include:

- Image classification

- Object detection

- Style transfer

Models lacking a non-maximum suppression layer, or those that use MultiArray for input/output data, are not supported.

Note: This app includes no model.

[For iPhone/iPad]

To use the app, first prepare a machine learning model in CoreML format using CreateML or coreml tools. Then, copy this model into the iPhone/iPad file system, which is accessible through the iPhone's 'Files' app. This includes local storage and various cloud services (iCloud Drive, One Drive, Google Drive, Dropbox, etc.). You can also use AirDrop to store the CoreML model in the 'Files' app. After launching the app, select and load your machine learning model. 

Models can be accessed and opened directly from the 'Files' app using the export menu.

You can choose the input source image from:

- Video captured by the iPhone/iPad's built-in camera

- Still images from the built-in camera

- The photo library

- The file system

For video inputs, continuous inference is performed on the camera feed. However, the frame rate and other parameters depend on the device.

[For Mac]

Unlike the iPhone version, the macOS version does not support loading images from the Photos library or local documents. It supports live video input only.

When an external video input device is connected to your Mac, it will be used in preference to the built-in camera. If no external device is available, the MacBook’s FaceTime camera will be used.

The level meter displays the processing time for each frame. The gauge reaches its maximum at 30 ms. This is because processing times of 33 ms or longer will cause frame drops when running at 30 FPS.

[Vision LLM]

Switch to the Vision LLM tab in the Settings window to configure the connection, including the destination IP address.

The application is designed to work with the API provided by LM Studio. Since it uses an OpenAI-compatible API, it can also connect to other compatible services. iOS version also supports OpenAI and Anthropic with presets. Prepare your own API Key to access OpenAI and Anthropic API. Host and port setting accepts these expressions; localhost:1234,  192.168.1.1:1224, my-mac.local:1234. Note that default port of LM Studio is 1234.

The image analysis behavior is determined entirely by the prompt. For example, you can instruct the model with prompts such as “What is shown in this image?” or “Identify the location.”

When Notification Center integration is enabled, the application can send notifications when specified conditions are met. The notification conditions are also defined in the prompt. For example, you can include an instruction such as “If a person is detected, send a Notification Center alert.”

• Starting with version 3.1, notifications will be delivered to all devices signed in with the same Apple Account.

This application does not collect or transmit any personal information.

This application does not use any identifiers that can be used to identify a specific device or individual.

This application does not require an Internet connection unless requested by the user.

This application uses photos or images stored on the user’s device.

This application uses the device’s camera functionality but does not save any images captured with the camera.

This application transmits images only to a server specified by the user.

Effective Date: October 1, 2022

Last Revised: July 1, 2026