Get startedModels · Interactive validation
Test a model interactively
Run representative visual inputs through a selected model before committing it to a workflow or release.
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
Understand predictions, confidence, latency, and obvious failure modes for one exact model version.
Start here
What you need to know first
Check the inputs below before starting. If you are new, begin with the first-project guide. 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
- — A pinned model version
- — Representative visual inputs and inference parameters
Expected output and limits
- — Rendered predictions and raw structured output
- — Latency and runtime diagnostics
Core workflow
Choose media
Open Build & evaluate → Analyze images. Use the task switcher in the header: Analyze an image, Test & compare models, or Process images & video. Test & compare models supports one model, side-by-side comparison, and image batches. Upload an image or video. Detection is ready for an automatic first pass. For segmentation, name the objects you want outlined. Choose Ask about the image for a vision-language question, scene explanation, or text reading. Select a verified image-capable connection, enter a question, and read or export the answer. Connect the runtime in Connections first; catalog foundations alone are not runnable. The provider may charge your account. Answers are saved with the question and are not annotation boxes, masks, or measured accuracy. Open Advanced settings only when you need to choose an exact specialist model version.
Provide an input
Run the automatic first pass and inspect every returned box or mask. Video preview preserves source timing and source audio when present. Open Compare & batch for dataset browsing, exact model selection, comparisons, and image batches.
Inspect the result
Review boxes, masks, classes, confidence, and latency. Choose Review & correct to open the saved image and predictions as editable annotations in Annotation studio.
Continue deliberately
Add the model to a workflow, evaluate it formally, or return to the registry.
What this surface supports
Specialist predictions
Task-specific output from compatible runtimes
Confidence and latency
Structured output inspection
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
- — Confidence and output validity
- — Cold and warm latency
- — Qualitative behavior on edge cases
Common failure modes
- — Mistaking a hand-picked demo for evaluation
- — Testing different preprocessing than production
- — Ignoring low-confidence or abstained outputs
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
- — Interactive tests do not replace evaluation
- — Version identity stays visible
- — Inputs are not silently added to training data