Changelog

New features

Prompts is now generally available

Prompts is now available in Label Studio Enterprise SaaS.

Prompts is a tool that leverages LLMs to automatically generate predictions for your labeling tasks, allowing you to pre-annotate data and quickly bootstrap projects. It can also be used to evaluate your LLM prompts against ground truth data.

For more information, see the Prompts documentation and Accelerate Labeling and Model Evaluation with Prompts—Now Generally Available.

Pause an annotator

There is a new action to pause annotators. This is available from the Members dashboard and via the API.

For more information, see Pause an annotator.

Personal access tokens

There is a new type of token available for API access. The new tokens use JWT standards and, unlike the current tokens in use, allow you to set an expiration date.

You can enable or disable these tokens from the Organization page. Once enabled, they will be available for users to generate from their Account & Settings page. Legacy tokens can still be used unless disabled from the organization level.

For more information, see Access tokens.

Deep link annotations and regions

You can now link directly to specific annotations or regions within an annotation. These actions are available from the labeling interface in the overflow menus for the annotation and the region.

Feature updates

Set ground truths by user

There is a new action from the Data Manager that allows you to mark the annotations submitted by a specific user as ground truth annotations.

Breaking changes

Users in the Manager, Annotator, and Reviewer role are no longer able to retrieve their API access tokens from the Label Studio UI.

This does not affect the functionality of any existing scripts or automations currently using their access tokens.

Bug fixes

• Fixed an issue that was causing intermittent loading errors in the Data Manager.
• Fixed an issue where filters were not being respected when performing bulk annotation actions.
• Fixed an issue where bulk annotation was sometimes failing with a 500 error.
• Fixed an issue where bulk annotation was not respecting the Allow empty annotations setting.
• Fixed an issue that caused incorrect video frame to be rendered when pausing.
• Fixed an issue where users were getting their role reset if they were provisioned via SCIM without an assigned group.
• Fixed an issue where an error was sometimes thrown when loading the workspaces list.
• Fixed an issue that would cause an API error when switching workspaces and fetching the incorrect page of projects.

New features

Bulk labeling

There is a new Bulk label action available from the Data Manager. You can use this to quickly label tasks multiple tasks at once.

This feature also includes enhancements to the Grid View in the Data Manager. Now when viewing images, you can zoom in/out, scroll, and pan.

For more information, see the Bulk labeling documentation and Bulk Labeling: How to Classify in Batches.

Enable AI features

There is a new toggle on the Billing & Usage page (only available to users in the Owner role). You can use this to enable AI features throughout Label Studio. See AI features.

Feature updates

Updated Billing & Usage page

The Billing & Usage page (only accessible to users in the Owner role) has several new options:

• Early Adopter - Opt in to new features before they're generally available.
• Enable AI Features - Enables AI helper tools within Label Studio. See AI features.
• White labeling - Contact sales about enabling white labeling.
• Custom Scripts - Contact sales to enable custom scripts. See Custom scripts for projects.

Anthropic model support for Prompts

You can now configure Prompts to use an Anthropic model. For more information, see Model provider API keys.

Bug fixes

• Fixed an issue where the Annotator Performance drop-down was not filtering the results as expected.
• Fixed an issue where users were unable to select and move bounding box regions after adding brush regions.
• Fixed an issue where seeking within a video would display duplicate frames in the the Video tag.
• Fixed an issue when managers could review skipped tasks in Quick View.

Feature updates

Gemini and Vertex AI model support for Prompts

You can now configure Prompts to use Gemini and Vertex AI models. For more information, see Model provider API keys.

Added OpenAI o3-mini model to Prompts

When you add an OpenAI key to Prompts, you will now see the o3-mini model available in the Base Model drop-down menu. For a list of all available models, see Supported base models.

Display the prediction score in the labeling interface

When you have annotations generated by predictions (pre-annotations), you will now see the prediction score (also known as the "confidence score") under the model name in the labeling interface tabs.

Updated template images

The thumbnail images for the pre-built templates have been redesigned.

Clearer description of annotation count

The annotation counter at the bottom of the Data Manager has been updated to read "Submitted Annotations." It previously read "Annotations," which could cause confusion.

Region number in Relations panel

The Relations panel now displays the number identifier for the region when viewing relations between regions.

Improved readability for relation values

If you are using <Relation> tags in your labeling config, you can specify a value describing the relationship between regions.

In the past, it could be difficult to read these values when there were several overlapping relations and regions. With this release, the text value for the relationship is pulled forward into focus when you hover.

Security

Security enhancements and optimizations.

Bug fixes

• Fixed an issue where project to groups mapping was not working correctly for SSO.
• Fixed TypeError issues causing page crashes when using Prompts.
• Fixed an issue where the error "string indices must be integers, not 'str'" would appear when running Prompts with custom models.
• Fixed an issue that caused would sometimes cause project creation to fail when pasting code into the code editor.

Feature updates

• Previously, when exporting data in YOLO, YOLO_OBB, or COCO format, the images themselves were not included in the export.
  
  To improve this, we have introduced three new choices to the export options:
  • YOLO_WITH_IMAGES
  • YOLO_OBB_WITH_IMAGES
  • COCO_WITH_IMAGES
• Optimized the Projects page for faster load times.
• The Video Frame Classification template is now available when browsing templates within the Label Studio project settings.
• There is a new hotkey available for pausing and starting audio: ctrl+p/command+p. This is in addition to the space hotkey that performs the same function. However, this new hotkey is useful when you are working with audio and have a text area field in focus.

Bug fixes

• Fixed an issue with overlapping relations on the overlay on highlighting

New features

Image support for Prompts

You can now use image data with Prompts. Previously, only text data was supported.

This new feature will allow you to automate and evaluate labeling workflows for image captioning and classification. For more information, see our Prompts documentation.

Beta features

Use AI to create projects

We’ve trained an AI on our docs to help you create projects faster. If you’d like to give it a try early, send us an email.

Feature updates

Use Taxonomy for labeling

There is a new labeling parameter available for the Taxonomy tag. When set to true, you can apply your Taxonomy classes to different regions in text. For more information, see Taxonomy as a labeling tool.

Bug fixes

• Fixed an issue where the Label Studio version as displayed in the side menu was not formatted properly.
• Fixed an issue where, in some cases, project roles were reset on SAML SSO login.
• Fixed an issue affecting Redis credentials with special characters.
• Fixed an issue where task IDs were being duplicated when importing a large number of tasks through the API.
• Fixed an issue where users were not being redirected to the appropriate page after logging in.

New features

Paginated multi-image labeling

Paginated multi-image labeling allows you to label an album of images within a single task. When enabled, a page navigation tool is available within the labeling interface.

While you can use paginated multi-image labeling with any series of related images, it can also be especially useful for document annotation.

For example, you can pre-process a PDF to convert it into image files, and then use the pagination toolbar to navigate the PDF. For more information, see our Multi-Page Document Annotation template.

To enable this feature, use the valueList parameter on the <Image> tag.

Classification reports for Prompts

When your Prompt includes classification elements, you can now view a report that tells you how often each class was identified. This is available for the following tags:

• Choices
• Labels
• Pairwise
• Rating

Feature updates

• Several usability enhancements for Prompts, such as a descriptive header and a reordered version menu.
• Improved the inter-annotater agreement API so that it is more performant and can better handle a high number of annotators.

Bug fixes

• Fixed an issue where SSO/SAML users were not being redirected back to the originally requested URL.
• Fixed an issue where a timeout on the inter-annotator agreement API would cause missing data in the Annotator Summary table on the Members page.
• Fixed an issue where the default date format used when exporting to CSV was incompatible with Google Sheets.
• Fixed an issue where users were unable to edit meta information that they previously added to bounding box regions.

New features

Set task reservation time

There is a new project setting under Annotation > Task Reservation.

You can use this setting to determine how many minutes a task can be reserved by a user. For more information, see Project settings - Task Reservation.

By default, the task reservation time is set to one day (1440 minutes). This setting is only available when task distribution is set to Auto.

Feature updates

• Updated the placeholder text that users see in the labeling configuration preview.

Bug fixes

• Fixed an issue where commas in comment text breaking were causing errors when exporting to CSV from the Annotator Performance report.
• Fixed an issue where annotators were seeing a tooltip message stating that the project was not ready yet, even though the project had already been completed.
• Fixed an issue that was causing 404 errors in the Activity Log.
• Fixed an issue where users were unable to deselect tools from the toolbar by clicking them a second time.
• Fixed an issue where users were presented with Reviewer actions even if the annotation was still in Draft state.
• Fixed an issue with the Source Storage editor in which some fields were overlapping in the user interface.
• Fixed an issue with the Data Manager filters when the columns are different from those in the labeling config and when $undefined$ is present in the task data.
• Fixed an issue where filter options in the Data Manager would disappear on hover.
• Fixed an issue which caused XML comments to incorrectly be considered in the label config validation.
• Fixed an issue causing an error when marking a comment as read.
• Fixed an issue where an error message would appear when selecting or unselecting the Get the latest news & tips from Heidi option on the Account Settings page.

New features

Flexible label configuration support for Prompts

Previously, the projects that you could use with Prompts were limited to relatively simple labeling configurations (for example, only having one set of Choices or one set of Labels).

With this change, you can now use complex projects with Prompts. For example, project with multiple Choices, Labels, and a combination of both. It also adds support for TextArea and Pairwise control tags.

Also as part of this change, you will no longer be asked to select a project Type when creating a new Prompt.

For more information, see Create a Prompt.

Export information from the annotator performance report

There is a new Export action available from the annotator performance dashboard.

• Report - Download the information in the dashboard as CSV or JSON.
• Timeline - Download a detailed timeline of all the user's annotation actions within the timeframe, including when the began and submitted each annotation.
• Comments Received - Download a CSV file with all of the comments that other users have left on the user's annotations.

Feature updates

Edit regions when classifying video frames

We recently introduced the ability to perform video frame classification with the <TimelineLabels> tag.

You now have the ability to edit the frame spans you select in the timeline editor, making it easier to control which frames you want to label.

Improved usability on project settings pages

There are a number of project settings that are only applicable when auto distribution is enabled for users.

To prevent confusion, settings that are not applicable will be hidden when manual distribution is enabled.

This means the following settings will be hidden when Annotation > Distribute Labeling Tasks is set to Manual:

• Annotation > Task Sampling
• Quality > Overlap of Annotations
• Quality > Low Agreement Strategy

Improved TextArea look and feel

TextArea elements have been updated to reflect the look and feel of other labeling elements.

Improved error messages when using Send Test Request for connected ML backends

When using the Send Test Request action for a connected ML backend model, you will now see more descriptive error messages.

Bug fixes

• Fixed an issue in which in certain scenarios users were unable to receive a password reset email.
• Fixed an issue where project-level roles did not affect role upgrades performed at the Organization level.

New features

Use an LLM to help you draft Prompts

There is a new Enhance Prompt action available when writing a prompt.

You can use this feature to allow an LLM to enhance the prompt you already have, using your initial prompt and the tasks as context. For more information, see Draft and run prompts.

Feature updates

Estimate cost for Prompts

You will now see an estimated cost before running a prompt. This estimate is based on the number of tokens required:

Bug fixes

• Fixed an issue where links were not resolving when using multiple S3 storages.
• Fixed an issue where non-unicode symbols would cause the Activity Log page to not load.
• Fixed an issue that could produce duplicate accounts when synced from SCIM.
• Fixed an issue where instructions were not visible to reviewers in the Review Stream.
• Fixed an issue preventing linking comments to an audio field.
• Fixed an issue preventing linking regions in audio tasks.
• Fixed an issue where deleted annotator users were not available as option when building filters in the Data Manager.

New features

Link comments to fields

You can now link comments to specific regions or fields within an annotation.

This change will help improve clarity between Annotators and Reviewers, enhancing the quality review process.

For more information, see Comments and notifications.

Flexible reject options for reviewers

There is a new option on the Review page of the project settings: Allow reviewer to choose: Requeue or Remove.

When enabled, reviewers will see two reject options for an annotation: Remove (reject and remove the annotation from the queue) and Requeue (reject and then requeue the annotation).

This is useful for situations in which some annotations are incorrect or incomplete, but are acceptable with some fixes. In those cases, it may be more useful to send the task back to the annotator. In other cases, where the annotation is far from correct, it is better to remove it entirely.

For more information, see Project settings - Review.

Custom models for Prompts

You can now add a custom, self-hosted LLM to use with Prompts. This will allow you to use your fine-tuned models with your Prompts workflow.

Custom models must be compatible with OpenAI's JSON mode. For more information, see Model provider API keys.

Version control for Prompts

There is a new Save As option when saving a prompt. This allows you to save named versions of a prompt as you refine it.

You can select between saved versions to compare their metrics. You can also manage older versions by updating them, deleting them, or renaming them.

Feature updates

Enhanced Prompts evaluation metrics

When you run a prompt, you will now see the following metrics:

• Inference cost - The cost to run the prompt evaluation.
• F1 Score - Precision and recall scores for the prompt. (Ground truth data sets only)

For more information, see Draft and run prompts.

Hotkey to show/hide all regions

A new hotkey (Ctrl + h) has been added. Use this shortcut to hide all regions. Or, if no regions are visible, show all regions.

Bug fixes

• Fixed a small UI issue seen when displaying drop-down menus with multiple nested selection options.
• Fixed an issue where users were shown a 500 error when attempting to create a project without first selecting a workspace.
• Fixed an issue where the application would crash when annotators who have also had a project role of Reviewer would navigate to the Data Manager.

Feature updates

Multi-class text classification in Prompts

Prompts, our Enterprise tool for evaluating and refining LLM prompts, now supports multi-class text classification projects. This will enable you to label your text with more than one choice, allowing for more complex classification use cases.

This also means that projects with the choice="multiple" parameter set will now appear in the Target Project drop-down menu when creating a Prompt (assuming that the project meets all other eligibility criteria).

Bug fixes

• Fixed an issue where the instructions modal was not appearing for reviewers even though Show before reviewing was enabled in the project settings.
• Fixed an issue in which draft lead time could incorrectly inflate the lead time calculation for an annotation.
• Fixed an issue in which Data Manager drop-down menus were inaccessible in smaller viewports.
• Fixed an issue which caused the Label Config UI preview to display stale information.
• Fixed an issue in which the agreement score was not updating after a reviewer completed the Fix+Accept action.
• Fixed an issue where images were improperly spaced in the Annotator Agreement Matrix.
• Fixed an issue in which Label All Tasks would not respect filters that had been applied in the Data Manager.
• Fixed an issue in which users were not able to use Quick View to load tasks that included an external taxonomy.
• Fixed an issue that caused the time portion of the DateTimePicker in the project Dashboard to not display correctly.

New features

Video frame classification

You can now apply labels to video frames. Previously, we only supported per-video classification.

This new feature allows you to apply labels at a per-frame level. You can implement this feature using a new tag: <TimelineLabels>.

For more information, see New! Video Frame Classification.

Named Entity Recognition projects for Prompts

You can now select NER projects when creating a Prompt. Previously, Prompts only supported Text Classification projects.

For more information, see Named entity recognition (NER) in our Prompts documentation.

Feature updates

• When creating regions that have start and end times (such as when annotating sections of an audio track), you will now see the duration of your selection under the Info tab.

• The Projects page header has been updated with a more compact design:
  
  - The search bar, projects drop-down menu, and create actions have all been consolidated onto one line.
  
  - The Use Template action has been moved and is now available as a drop-down option within the Create Project button.
  
  - For better clarity, the Use Template action has been renamed Create from saved template.
  

• You will now see a progress bar when performing searches against the activity log to indicate that the search is still processing.
• For better clarity, the Sandbox workspace has been renamed Personal Sandbox.

Bug fixes

• Fixed a regression issue with BigInteger support in the Data Manager.
• Fixed an issue with the Activity Logs page where some options would become unavailable.

Feature updates

• Changed the default behavior of the Project Dashboard so that it no longer defaults to the “include time” option in the calendar.

Bug fixes

• Fixed an issue in which project-level roles could not be reverted once set.
• Fixed an issue where user limits were not being enforced when users were added via LDAP.
• Fixed an issue where setting a task agreement threshold incorrectly affected counts in the project Dashboard.

New features

JavaScript for labeling customization

You can now add JavaScript to your Label Studio projects to further customize your labeling experience.

Note that due to security precautions, custom scripts must be enabled for your organization before will see the Scripts option when configuring the labeling interface. Contact your account representative to request this feature.

For more information, see the following resources:

• Custom scripts documentation
• HumanSignal Blog - Transform Your Labeling Workflow with Custom Scripts
• Video demo - Announcing Custom Scripts for Label Studio

Feature updates

You can now use Azure OpenAI when creating Prompts. For more information, see Model provider API keys.

Bug fixes

• Fixed an issue where users were seeing a runtime error when loading the dashboard in situations in which the project’s labeling configuration did not include labels.

Feature updates

• There is a new Performance Score column on the project Members page.
  
  The Performance Score column reflects the overall performance of annotators. This score takes into account all review actions, including annotations that were initially rejected and later accepted. The calculation is as follows:
  
  - Each annotation review action (Accept, Reject, Fix+Accept) contributes to the score.
  
  - The score is calculated by summing the scores of all review actions and dividing by the total number of review actions. For example:
  
  - If an annotation is rejected twice and then accepted once, the Performance Score would be (0 + 0 + 1) / 3 = 33%.
  
  - If an annotation is rejected once and then fixed+accepted with a score of 42%, the Performance Score would be (0 + 0.42) / 2 = 21%.
  
  This is different from the Review Score, which only reflects the current accepted/rejected state of annotations.

• When you make changes to the labeling configuration and attempt to navigate away before leaving, you will now see a warning message prompting you to save your changes.

• You can now reorder tabs in the Data Manager by dragging and dropping them.

Bug fixes

• Fixed an issue with breaking Text/HyperText content that contains emoji when regions are added.
• Fixed an issue where the project summary was not being included when duplicating a project.
• Fixed an issue where tracing was breaking presigned URL requests.
• Fixed an issue where users were seeing an error when visiting the Annotator Performance Report page without first selecting a user. Instead, users should be redirected away from the page until a user is selected.
• Fixed an issue where workspaces could be deleted through the API even if they still contained projects.
• Fixed an issue where, when attempting to select multiple bounding boxes by pressing Command (or Ctrl), a new bounding box would be created instead.
• Fixed an issue with region labels were not displaying the index number if there was not label text.

Feature updates

• When viewing regions with the Show region labels option enabled, the region label will now include the same index identifier that you see in the regions list.
  
  Before:

After:

• When using the Annotator Performance Reports, you can now select whether you want to aggregate data by creation date or updated date.

Bug fixes

• Fixed an issue with agreement scores for Annotators being inaccurately calculated on the Members page within projects. This issue would appear when a Reviewer would reject an annotation and then later accept it.
• Fixed an issue where images were improperly resizing after loading due to how the original dimensions were set.
• Fixed usability issues seen in the Members table once the member user list grows large enough.
• Fixed an issue with displaying large integer numbers in the Data Manager.
• Fixed an issue with tooltips that was causing errors in the transition effects.
• Fixed an issue where Annotators where able to resolve comments made by Reviewers, when this action should not be available to them.
• Fixed an issue where annotation history was not working correctly if created_by was null.
• Fixed an issue affecting SAML users caused by changes to the HumanSignal app domain.
• Fixed an issue where the docs link icon was not properly formatted.

Bug fixes

• Fixed an issue where users were unable to export in Pascal VOC XML format when applying bounding boxes to images.
• Fixed an issue where usernames in the Annotator Performance Report were not displayed correctly if the user had a long email address.
• Fixed an issue where annotation history was not working correctly if created_by was null.
• Fixed a sizing issue affecting the icons for workspace actions.
• Fixed an issue where after an annotation had been fixed and accepted, the update action was not recorded when exporting the annotation history to JSON.
• Fixed an issue in which sometimes the Submit button would be displayed when it should be the Update button.

New features

Introducing the Prompts tool

The Prompts tool leverages ChatGPT to help you evaluate and refine your LLM prompts. You can also use Prompts to generate predictions for automating your labeling process, and to quickly bootstrap labeling projects.

For more information, see Automate Data Labeling with HumanSignal and our Prompts documentation.

Refreshed Label Studio interface

The Label Studio UI has been upgraded with updated colors and fonts, giving it a sleek new look while maintaining the same intuitive navigation you're familiar with. All Label Studio tools, features, and settings are still in the same place, ensuring a smooth transition.

Feature updates

Improved performance on the Projects list page due to improvement on the API level.

Breaking changes

Fixed an issue with Google Cloud Storage when the connection has the Use pre-signed URLs option disabled. In these situations, Google was sending pre-signed URLs with the format https://storage.googleapis.com rather than sending BLOBs.

With this fix, Google Cloud Storage will begin returning BLOBs/base64 encoded data when Use pre-signed URLs is off. This means that Label Studio will start reading data from Google Cloud Storage buckets, which can result in large amounts of data being sent to your Label Studio instance - potentially affecting performance.

Bug fixes

• Fixed an issue where Redis storage connections were causing errors due to a missing field in the storage form (Storage Title).
• Fixed an issue where annotators were not being prompted to leave a comment when skipping a task, even though the project settings required them to do so.
• Fixed an issue where connected ML backends were unable to return more than one prediction per task.

Feature updates

When using the annotator performance report, you can now filter by project or workspace.

Bug fixes

• Fixed an issue where sometimes actual usernames were being replaced by a generic “Admin” username in the annotation history.
• Fixed an issue which caused the annotator dashboard project selector to not initialize on first interaction.
• Fix Redis connection with SSL.

Feature updates

• When you create a new snapshot, you will now have the option to include annotation history:

• For organizations with the HIDE_STORAGE_SETTINGS_FOR_MANAGER environment variable set to True, Managers will now be able to sync data from external storage as necessary rather than request assistance from an Admin user.

Bug fixes

• Fixed an issue where regions were disappearing when a user would switch between annotations and predictions.
• Fixed an issue where the Batch Predictions action was failing under certain conditions.
• Fixed an issue where tooltips on the dashboard were causing confusing mouseover behavior even when not visible.
• Fixed an issue with button sizing inconsistency when selecting date ranges.
• Fixed an issue that prevented users from fetching the URL for TimeSeries objects.