Getting Started with Screph

Installation and first launch

1. On the download page, choose an available published file and follow the installer's instructions. A particular build may include a different set of features. Back up important projects and settings before updating.
2. Start Screph and open the workspace window. If the application is in the system tray, use its Open Screph command. Opening the workspace and taking a screenshot are separate actions: choose a source in the next step.
3. In General Settings, check the language, theme, and application mode. Keep a clear tool profile and hints enabled. The settings map shows where to find the remaining options.

Optional components are prepared for the selected task: dependencies and updates for processing and training, and LLM and speech connections for model requests and voice descriptions. The initial manual annotation route does not require an external model request.

Install the FakerInput system driver and Arduino firmware separately when using the corresponding input method. See Windows and Permissions for setup.

Mode and task

• General: image annotation and general computer vision operations. Start here for a photograph or for exploring regions in a single frame.
• GUI Automation: prepare interface data for a separate automation script. An image of the target window is enough to start; mouse and keyboard control is configured separately.
• Industrial and UAV are experimental subject workflows without production-ready control.

The mode determines available annotation types and operations. Save the project before switching; see Application Modes for status and limitations. With video, learn one frame first, then move to ranges and sequences.

Workspace window and new project

The Selector canvas shows images and annotations. The toolbar selects the action, the information panel shows the selected object’s properties, and the bottom tabs open Capture, Computer vision and other workspaces. If a panel is hidden, restore it through the View menu.

1. For a separate task, click New Project on the toolbar. If the current project contains data, choose to save it, continue without saving, or cancel. A new project clears the previous annotations and frame state, so do not use this action to add another object.
2. Check the selected tool and read the hint below Canvas. It helps distinguish selecting an object from creating an area, moving, and resizing. For a convenient overview, use View → Fit Image.

Source and selected frame

1. Open Capture and, under Source, choose files for the first image. Add your own file and wait for loading to finish: the image should appear on Canvas. For a real recording, choose a screen, window, monitor, camera, or stream and check the source settings before starting.
2. For a video or frame folder, select the desired frame on the timeline. Playback helps locate the moment; stop it before annotating, and check the frame number and image. Recording, playback, and a single snapshot are different actions.
3. Make sure the object is fully visible, the image is legible, and the material belongs to your task. If another source is needed, replace it before annotating. A sequence range is set separately; selecting a frame does not select all frames.

Recording, storage settings, and frame export are covered in the Capture guide. Changing sources and working with video require checking how objects are linked to images and frames.

First area and object properties

1. On the toolbar, choose a regular area with a rectangular shape. Draw around the first object in the image. A box appears on Canvas, and the selected area becomes available in the information panel.
2. Give the area a clear name and description: what it contains and what role it plays in the task. For example, Confirmation button and Confirms the completed form. Check the object type; name, shape, and description serve different purposes.
3. Switch to selection and editing. Adjust position and size to the object's boundary, then create a second area. Check the selected object in the properties panel after each action. Undo an incorrect edit with Edit → Undo.

You can enter a text description manually. Voice annotation requires configured speech recognition; review the text after recording and before saving. A manual description is enough for a first project.

Feature and processing area

An element represents a task object; a feature describes its visual property or part. When a detail matters, select its owner and create an appropriate feature, such as an area containing text or a distinctive part of an object. Check the feature's owner, source, and boundaries in its properties.

An area of interest defines the material for processing. A regular area on the current image is enough for a first calculation. Do not assign a hierarchy merely because one box is drawn inside another: relationships are explicit.

Relationships between objects

1. If two objects are related conceptually, use the relationship tool and select its participants. Add a description for a domain relationship; for a hierarchy, check the parent and child.
2. Select the resulting relationship and check its type and both objects. Open the graph in the information panel to see its structure. One verified relationship is enough for a first project; geometric proximity alone does not create one.

Shapes, features, groups, relationships, and descriptions are covered in the annotation guide. You can explore these detailed operations after saving your first project.

One operation and a result check

Add processing if the task needs masks, edges, text, or matching. Manual annotation does not require it. To get started, use one available local operation on the selected area, such as edge detection.

1. Select an area and open Processing. Check the source image and input boundaries, choose an operation, and inspect its parameters. If a component is unavailable, prepare the listed dependency first.
2. Start a preview and compare the source with the result. Change one parameter and recalculate to understand its effect. Check the selected output: an image, mask, and table can convey different information.
3. If needed, save and attach the result to the project. If an operation proposes an annotation change, check geometry, object type, and duplicates separately, then confirm a supported apply action. Inspecting and saving a result do not themselves accept the proposal.

In Simplified mode, configured processing for a regular region runs and saves automatically. The explorer shows calculation progress, errors, outdated results, and completed saves. Check the result; a proposed annotation change must be accepted separately.

After changing the source, frame, or parameters, run the calculation again. Chains, graphs, and sequence processing are covered next in the computer vision guide; first check one operation on one source.

Save and reopen

1. Check the source, boxes, names, descriptions, relationship participants, and required processing results. Open File → Save Settings… to check the project location and selected profile. For a first project, choose a profile without automatic agent execution if coding work is not yet needed.
2. Save the project with File → Save Project or Ctrl+S. Wait for writing to finish and read the result message. Saving the main project and subsequent actions in the selected profile have separate outcomes: an IDE handoff error does not establish that saved data was lost.
3. Open File → Load Project or press Ctrl+O. In the manager, select a saved project, check its location and details, then load it into Selector. Confirm that the image, annotations, relationships, and attached materials are available again.
4. To back up a project, save the entire project directory with its images, masks, and related files. After moving it, open the copy from its new location and check its contents; the main file alone does not replace linked materials.

The manager also shows autosaves and read errors. Before using Restore Autosave, make a copy of the current project and check the selected record. Autosave can help recovery, but transfer and backup require a verified project directory.

Next task route

• Working with code. Screph Code is a built-in agent IDE. After saving the project, choose a separate working folder, prepare a model connection, and review the materials sent with the request. State the program's expected behavior and constraints, then review the response and file changes. The guide describes the editor, chat, executor, and access mode. Saving the project, passing context, requesting an agent response, and running the resulting code are separate steps.
• Dataset. In Datasets and Training, create a dataset with a task and classes, then add files, Canvas material, or selected Capture frames. Review the samples and assign annotations. Import is explicit: changing a box in the main project does not update an already saved dataset sample.
• Training. After checking the data, choose a model, parameters, and a prepared environment. Start Training first prepares a preview of real samples; confirm the run after reviewing it. Then inspect metrics and errors. Model files, release, and permission to pass a model to the final program require separate setup.
• Execution and input. For GUI Automation, prepare an Automation Runtime script, target window, and input method separately. Before running, review the program's actions and the selected executor's permissions. Windows and permissions explains system integration. An agent response, annotation, or computer vision result does not establish that the final program works.

The detailed route through datasets, training, and Screph Code is in the user guide. A Screph project, dataset, coding workspace, and execution results are saved as separate materials; check each result you need before finishing.

Tutorial demonstrations

1. In General Settings, find Learning Demonstrations, choose a topic or the full available list, and start a demonstration. Good starting points are sources and Capture, Canvas tools, relationships, and one processing operation.
2. Check the language and availability of prepared narration. Voice, hints, character, and learning-line settings are available in the demonstrations panel. If a recording needs preparation, wait for it to finish or cancel it.
3. Use pause, resume, and stop during a demonstration. After the learning route, repeat the action in your own project and check that it is saved. The character accompanies demonstrations; its settings do not add a permanent assistant to every window.

Most demonstrations open a separate learning window. A prepared agent response is labeled as prepared: playback is not a new model request. The examples on this page also use learning data; actions in them do not save the application's main project.

Offline help

In the Documentation tab, interactive help opens site pages with learning examples and requires a network connection. Saved Help works offline: it includes the contents and guides to getting started, modes, Capture, annotation, and processing. If the site is unavailable, the application switches to Saved Help. Help follows the application language. An update downloads a consistent set of pages and illustrations; on failure, the previous set remains. Saved Help presents interfaces as images.