Get startedStart · Product map
Understand Score Studio
Start with an image, a clip, or existing data. Learn what is saved, what runs automatically, and what needs your approval.
Source reviewed 2026-09-16. Availability depends on your installation, permissions, and compatible runtimes. A supported path is not a guarantee of model quality or production readiness.
Outcome
Know where to start, what Score Studio will do, and when it needs your review.
Start here
What you need to know first
Check the inputs below before starting. A dataset holds media and labels; a model produces results; a deployment makes a selected model version callable. Creating one does not create the others.
Bring these inputs
- — Uploaded images, a clip, an existing dataset, or a generation brief
- — Visible object names and permission to process the media
Expected output and limits
- — A saved analysis or a reviewed build setup, depending on your choice
- — A model and test report only after successful training and testing
Core workflow
Find your way around
Use the left sidebar for Projects, Library, and Production. More tools is expanded by default and can be collapsed manually. It contains Analyze images, Annotation studio, Training, Evaluations, Workflow builder, The Vision Lab, Agents, and Runs. Open Workflow builder to create a workflow or edit a saved one. Your name at the top of the sidebar opens the account menu: Settings, Connections, SDK & API, and Documentation. On smaller screens, open the navigation menu first.
Choose a task
Projects opens on dated rows of saved projects, each with its next action. New project opens the project creation form: name it, then add your data and model. The Get started page introduces the available tasks. Choose Build a model to train on your images, Label a dataset to prepare training labels, Analyze images or video to find objects or ask image questions, or Evaluate a model to check accuracy. Each option explains what you need and what you get. These are the same task destinations everywhere; choose the vision task and input in that workspace. Library holds reusable datasets and models.
Use managed image questions
Analyze images offers Fast and Ultra through Score-managed AI capacity. Free receives a small lifetime allowance; Smart and Pro receive a monthly allowance. Each successful request deducts its actual inference cost, and work stops when the allowance reaches zero. Smart and Pro can buy fixed capacity packs. Pro can instead use a supported external provider account owned by the workspace.
Inspect every image in one workspace
Standalone and project image analysis share one preview-and-controls layout. Detection, segmentation, classification, and visual questions use the same workspace structure. Keep the image visible on the left, choose the result type in the toolbar, and use the model and task controls on the right. Box and mask results keep zoom, original-image comparison, class filters, and object inspection. Run metadata is visible below the controls; Export results includes the full context and raw result. When no predictions are returned, the workspace shows one explanation and offers a new input or manual annotations; empty overlays and object-inspection controls are omitted. Saved results lead with review or project creation; use Result actions for export, another analysis, or alternate resource views. Unconfirmed results keep export directly available. Model selection stays in the task panel; version and confidence controls remain available. Changing the task or settings does not start inference. On phones, the panels stack with bounded scrolling for long results. Review editors show Identify objects in a consistent assistance panel with availability and prerequisite explanations. Save and Confirm actions remain in the task footer.
Inspect an image with a VLM
In Analyze images, choose Ask about the image. On desktop, the image stays beside the formatted answer and the question box remains below it. Choose Fast, Ultra, or a Pro-connected model, type a question or select a starter, and submit with the button or Ctrl/Command + Enter. Each question is independent. Copy or export an answer, or reopen it from Recent results. Billing details identify whether managed capacity or a customer provider pays for the request.
Manage the complete project workflow
Open a project for a concise summary of its selected data, model, and next step. Use the Data, Models, Analyze, and Deployments page tabs for focused work. Completed builds link to their exact saved test reports; model choices remain visible until resolved. The overview shows selected resources, evaluation metrics, and experiment records with direct links to each. Open workflow leads to the typed graph built from the project's pinned data and model versions.
Bring media and name the objects
Choose Upload images or a clip, Use a dataset, or Generate a dataset directly inside Build a model. Score Studio prepares the source once, checks representative images, and brings proposed object classes and visible descriptions into the next step for you to adjust. Clips are sampled across the full timeline; a workflow is optional.
Review before training
Review build setup does not start training. Start building approves the steps. Confirm proposed annotations when asked; saving drafts alone does not confirm them. Hosted training requires a supported GPU provider account that you connect and verify with your own API key. Local training remains available in Community installations. Hosted automated training, segmentation, reward optimization, and distillation are paused; connecting a provider does not enable these local-only paths. Score Studio coordinates the run and does not supply GPUs.
Test, then decide
Follow progress in Overview or Runs. Inspect the saved model and test report. Selecting a model for a project does not deploy it; Production is a separate decision.
Replay the guided tour
Open the account menu → Settings. In Product tour, choose Replay tour, or Continue tour to return to a paused step. You can skip at any time; the tour never starts work or spends credits.
What this surface supports
Result-based onboarding with current feature breakdowns
Project lifecycle map with an editable visual workflow
Image, clip, dataset, and generated-data entry
Assisted annotation with explicit human review
Reviewed-data training and held-out testing
Separate deployment and monitoring controls
Expert section
Contracts, signals, and failure modes
Use this section when you are defining acceptance criteria, automating the surface, or reviewing whether its output is safe to promote downstream.
Quality and operating signals
- — Task completion and cost
- — Evaluation quality
- — Production conformity
Common failure modes
- — Starting with a model instead of an outcome
- — Deploying without a benchmark or feedback path
Expert release checklist
□ Inputs and dependencies are pinned to immutable versions.
□ Acceptance metrics include critical classes and operating slices.
□ Failure, retry, cost, and rollback behavior are understood.
□ The resulting artifact has an owner and a downstream review path.
Engineering safeguards
- — Release gates remain explicit
- — Inputs and outputs retain lineage
- — Supported long-running work stays visible in Runs