Dependencies and updates

Installation contents, additional environments, models, and external programs for Screph tasks. How to prepare components and check the function you need.

What is installed with the application

The Screph interface, image opening, annotation, and project saving require the main Screph files, Qt/PySide6 interface libraries, and core data-processing libraries. In the current Windows installation route, these parts are linked and installed as the application's base. Python does not need to be installed separately for the interface itself.

Installation part	Purpose and selection
Core files, interface, and core libraries	The application, resources, Qt/PySide6, image and data processing including OpenCV, NumPy, and Pillow, and libraries for the connections in use. This is the required installation base.
Screph components and dependencies	Tools for checking and preparing optional environments. The current Windows route includes materials for local Python, automation and coding environments, OCR, video, and local speech; selected settings determine which environments are installed.
Computer vision and ML	Additional machine-learning libraries. The separately installable ML environment package is needed by workflows using the selected external ML environment. Model weight files are checked separately.
Tutorial demonstrations	Example projects and prepared narrated demonstrations. A separate component offered by default. Selecting a character in settings does not install its files.
Screph Pro Agent	An optional component of the built-in Screph Code agent IDE. It can be updated separately from the main application; the selected agent also needs its corresponding environment and connection.

The table describes the current installation route. Actual contents of a downloaded installer and installed version depend on that release and selected components. You can check them in the maintenance tool and Components and Environments tab. Contents of a published build without confirmed component information remain unknown.

The small Windows installer may download components. Offline installation requires the complete set, with the package directory next to the installer. If your task uses an ML package and models, prepare them in advance.

Practice example. Changes apply only here and reset when you leave the example. Match the workflow to a component in the table. Check whether it is enabled, its selected source, and status; model files and external programs are checked separately.

Dependencies page

Install the program, prepare the runtime environment, and download model weights separately. A function may also require a model file, suitable hardware, and a configured connection. Its status shows why it is unavailable.

In the main window, the Components and Environments tab shows a summary and table with component enablement, selected source, status, selected and detected paths, size, and potential space savings. You can rerun checks, install or repair a local environment, and remove its local files. Installation starts only after a separate action.

Find the component for your workflow in the table and check whether it is enabled.
Choose a local installation or a compatible component already installed. For Python, select a specific environment with the required libraries.
Click “Check again”. Review the selected source’s status and any error reason.
If needed, choose Install or Repair Locally and wait for the result. Then check the model, device, or connection in the feature's settings.

An environment check may take time on its first import of large libraries. It checks the selected executable and required libraries but does not download models or start training. Checking an existing Python environment does not install or update its packages.

Local installation and existing environment

Local: Screph prepares its own component in the ScrephData data directory. This is the main Windows route; the files do not replace the system Python environment.
Found globally: use a compatible application or environment that is already installed. Use Found Global selects only components that passed validation.
Custom path: specify the exact executable. For Python, the existing-environments dialog shows each path, version, and reason for incompatibility. Advanced mode offers more path selection options.
Skip: the component is not used. Workflows that depend on it remain unavailable until a suitable source is selected.

Global installation is available only for supported components in advanced mode. It differs from using an already installed component. If the selected source is damaged or incompatible, Screph shows an error; choose another source explicitly.

Check free disk space using the size and potential savings shown in the table. The automation, OpenCV, and local speech environments may share local files. Before deleting them, stop running tasks and check which functions use the same environment.

Practice example. Changes apply only here and reset when you leave the example. Choose the component source and check its path. ML package options are set separately; after changing them, rerun the check and read the reason a component is unavailable before installing or repairing it.

What features require additions?

Function	What is required
Run Python programs and automation	Python and an automation environment with NumPy, Pillow, OpenCV, and PyAutoGUI. This is a separate environment for running programs; a working Screph interface does not establish that it is ready.
Classic computer vision	OpenCV and image-processing libraries. The main installation contains libraries for the application itself; an external OpenCV/CV environment is checked separately for runs that use it.
OCR / Tesseract	The Tesseract executable and data for the selected language. Current local installation materials include English and Russian data. Another language needs its own file; checking the Tesseract program alone does not establish that every language is available.
Video operations	FFmpeg for operations that use it, including precise trimming. Screph checks the exact path and whether FFmpeg can start; one available video decoder does not establish that every operation is ready.
YOLO, FastSAM, SAM, OmniParser	A compatible ML environment with the selected method's libraries, model files, and a suitable compute device.
Training and models for frame sequences	An ML environment with Screph components for training and sequence processing, dependencies of the selected model, and experiment data. YOLO trackers use LAP; SAM 2 is enabled separately. PyTorch alone does not establish that the environment is ready.
Local speech recognition	Vosk, PyAudio, a language model, and an available microphone. The current local installation materials include a small Russian Vosk model. Select another language together with a suitable model.
Screph Code / OpenHands	The Screph Pro Agent component, a coding environment, and a configured connection to the selected agent or LLM. An external editor also requires its application to be installed.
Codex / OpenCodex	Codex CLI is a separate optional component. OpenCodex is installed and started through a separate action in connection settings. Codex Desktop is installed independently; having the CLI does not install the graphical app.
ROS 2	Exporting a ROS program does not require ROS to be installed. Topic exchange, continuous streams, and running an exported node require a separate ROS 2 environment. Screph components do not include a ROS distribution.
External connections and devices	Network access and settings for the selected service; cameras and microphones require available devices and permissions. Arduino requires a device and suitable firmware; FakerInput on Windows requires a separate driver installation with UAC confirmation.

A ready-made machine learning environment does not mean that the desired model is already loaded. Models are tested separately; see “Computer vision models”.

ML package, libraries, and models

The ML dependency packages section sets the libraries for the selected ML environment. PyTorch performs neural network computations, TorchVision provides image operations, Ultralytics is used by YOLO/FastSAM, and LAP is used by YOLO trackers. Additional packages include timm, Transformers, einops, and Hugging Face Hub for supported models and downloads. Meta SAM 2 is selected separately and is off by default.

Library packages have dependencies: disabling one may make a selected method unavailable. The environment also needs Screph components for its tasks. When selecting an existing Python environment, all required imports are checked in the same interpreter.

You can prepare a local ML environment from a suitable ZIP package or an extracted directory. If a signed package is available to download, its version, target platform, size, signature, and contents are checked first. After downloading, select the archive for installation and wait for environment preparation to finish.

Network downloads are checked for checksums, signatures, and publisher trust. Installation checks package completeness and environment startup. If the signature is invalid or the platform is incompatible, choose a suitable package. For a local archive, verify the publisher separately: a checksum confirms only file integrity.

Select an available device for each CPU or GPU method. An NVIDIA GPU requires a compatible PyTorch build and system driver; installing the ML package does not install the graphics driver. Check the selected device and memory before running. Training also needs prepared data, experiment parameters, and space for results.

Local training does not require Ray. Training settings separately offer an experimental Ray executor for classification. The cluster must already have Ray Train, compatible PyTorch/TorchVision and Screph training components, an accessible shared folder, and result storage. A separate button checks Ray; a ready local ML environment does not establish that the cluster is ready. Detection and segmentation training through this executor are not yet available.

Practice example. Changes apply only here and reset when you leave the example. Match the selected YOLO model, its file, and compute device to the ML environment status. Prepare libraries and select weights separately.

Practice example. Changes apply only here and reset when you leave the example. Check the default executor and what happens to a local process when Screph closes. For experimental Ray mode, set up the connection and shared directories, then check the cluster separately. Choose experiment data and parameters before training.

Separate ROS 2 environment

ROS 2 is optional for developing an image-processing program. In Screph, open Settings and Tools → Automation → ROS 2. Opening the page and selecting a level do not start ROS.

Code generation: export supported saved CPU/CV processing as a ROS Python package and work with the result in Screph Code. Export does not require ROS to be installed or a model request.
Exchange: file import and frame export work without ROS; listing topics, receiving an image from a topic, and publishing require a connected ROS environment.
Continuous processing: receiving a stream requires a connection and an explicit start. Processing with a saved graph and publishing results are enabled separately.

For the connection, specify the absolute path to Python in an installed ROS 2 environment with rclpy, sensor_msgs, and std_msgs. That environment must have the ROS system libraries and selected DDS connection available. A general automation or ML environment cannot replace it. Screph does not install ROS automatically or add its libraries to its ordinary Python environment.

After selecting an environment, assess Apply and Connect by its connection result and topic list. A missing frame may mean the publisher has not started, or that a different domain, topic, or incompatible QoS was selected. Changed settings require applying and reconnecting. This connection does not support starting an environment through WSL, SSH, or a remote interpreter.

The exported ROS package contains an application program and saved processing, not a ROS distribution. Use ROS tools and the package instructions to install and run it on the target machine. Check the ROS environment, node packages, camera, or image source separately on that machine. ROS 2 Jazzy checks in a development environment do not establish that every installed Screph build is ready.

Before installing the component

Preparation may involve downloading and unpacking large files. Wait for the final status, not just the end of the download. If cancellation is still stopping background work, do not start another installation or delete its files at the same time.

Check available space and download source.
For a large machine learning environment, leave some space for archive, unpacking, and models.
If you are using a local package, make sure it is for your version of Screph and Windows x64.
Do not delete a shared component while other functions may be using it.

Program updates

Manual checks, the current version, channel, and automatic checks are available under Account → Updates and General Settings → Updates. Checking for and installing updates does not require signing in.

On Windows, the check retrieves version information for installed components. Updates are installed through a separate maintenance tool, including updates to individual components such as Screph Pro Agent. Complete installation in the maintenance tool, then restart Screph and check the version and component status.

You can enable or disable automatic update checks in General settings. The check retrieves information about available versions without sending the project, images or annotations.

Save your projects and settings before upgrading. After updating, recheck the components, models, and input methods that your workflow requires.

Application updates, ML package preparation, model weight downloads, and updates to an external ROS environment or editor are separate operations. An existing external Python environment is maintained by its owner; Screph checks do not update its packages.

If the maintenance tool is missing, information about available versions may still appear, but installing an update from Screph is unavailable. On Linux, use the update method stated for the specific package on the download page; the Windows maintenance tool and Windows environment packages do not apply.

Platform and system requirements

The managed installation of local environments described here is intended for Windows. In other distributions, choose the sources and packages offered by the installed program for your platform. Windows archives are not suitable for Linux.

Cameras, screen capture, microphones, and the selected input method may require operating-system permissions and device drivers. These are checked separately from Python libraries. Hardware input through Arduino requires suitable firmware and a selected device; FakerInput is installed separately from ML.

The Windows Long Paths button shows system status and lets you request a change separately. Windows may request UAC confirmation; restart Screph after enabling the setting. Inspecting components does not itself change the system setting.

If the test fails

Repeat the check after correcting a path or connection, or installing a component. Do not delete the entire data directory to resolve one error: save your project first and identify the specific missing or damaged component.

Output	Check and next action
Ready	The selected component passed validation. The model file, compute device, task data, and connection are checked separately.
Missing or incorrect path	Check the selected executable. Choose a suitable existing component or install one locally.
Library error or incompatible environment	Read the list of missing or failing libraries. Choose an environment with all required components; for a local environment, repair it from a compatible package.
Disabled, skipped, or unavailable on this platform	Check enablement and source. For an unsupported option, use an available delivery route; reinstalling a Windows package does not change the platform.
Not checked yet	Rerun the check after changing a source or path. An old result does not establish that the newly selected environment is ready.
Download, signature, or checksum error	Check the source, network access, and whether the package matches the platform and version. Download an incomplete or damaged archive again; a trust error cannot be treated as a successful installation.
Archive downloaded; feature unavailable	Check whether an archive was selected and the environment installation finished. Then check the models and feature settings.

Open the component details and read the reason shown.
Check the path, permissions, free space and network access.
Repeat the preparation or select the correct local package.
After installation, run the scan again from Screph.

In advanced mode, Diagnostics shows local parameters and the latest check result. This helps match a source and path to an error. Before sharing diagnostic materials, review their contents and hide personal paths and connection data.