# Test AI models online

## Test models before committing to an integration

U9 AI supports online testing for models exposed in its creation workspace and model-specific API integration tests. You need a registered account and sufficient credit for a generation. A guide or enabled price record alone is not proof that every upstream capability has been tested.

## A practical test workflow

1. Open [models and prices](/en/models) and choose the desired category and configuration.
2. [Sign in](/en/app), check available credit and open Create. If the desired model or mode is absent, [ask support](/en/support) before implementing it.
3. Enter a short prompt or upload a Markdown prompt. For image-to-video, provide a reference image using the model's supported input method.
4. In Advanced settings select supported duration, aspect ratio and resolution. Start with a small specification and review the displayed price before submitting once.
5. Save the request/task ID. Wait for a completed task with an output resource, then preview or download it. Reopen Request history for previous results.
6. Inspect [cost analysis](/en/app#costs) for final spending, known video seconds and processing time. Increase the specification only after reviewing output quality.

## Test credit and charges

Generation tests are billed at the displayed price and require available service credit. Contact [support](/en/support) with the model, intended scenario and expected test volume to ask about available test credit. A request is not an automatic credit approval; no free generation or fixed trial amount is promised.

## Validate APIs separately

Start with the read-only /v1/models check from [quickstart](/en/docs/quickstart). Then use a small generation to test the actual protocol, streaming and tool calls needed by your client. For coding tools, follow the [individual MD guide](/en/docs/agent-files). Do not blindly resubmit after a timeout; check existing task state and billing records first.

## What counts as a useful result?

Record model ID, mode, prompt version, duration, resolution, task ID, output quality, processing time and settled cost. Compare the same settings. Landing-page artwork is illustrative and is not a benchmark or a claim of guaranteed output quality.