AI video aspect ratio test: Check requested ratio, file dimensions, and final crop

An AI video aspect ratio test should inspect three boundaries: what the request asked for, what the downloaded file contains, and how the final player displays it. A valid ratio setting does not guarantee that a downstream crop preserves the subject or that a layout uses the file without distortion.

Confirm the requested configuration

Save the explicit model and aspect_ratio in the approved request. The current Seedance schema accepts 16:9, 9:16, 1:1, 4:3, and 3:4. Kling and Veo accept 16:9, 9:16, and 1:1. An unsupported ratio is a schema issue rather than an editorial crop problem.

Do not put aspect ratio only inside a structured prompt string. The route maps the actual body parameter; it does not extract typed configuration from arbitrary creative text.

If aliases are both present, the snake case value wins when non-nullish. Review the serialized request so a stale camel case setting does not mislead the operator about the selected ratio.

Inspect the actual file

After retrieval, inspect the decoded video dimensions and any relevant display-orientation metadata using your normal media tooling. Calculate the resulting display ratio from the file evidence. Do not infer dimensions solely from the requested resolution label or the player window.

A successful job and a video URL do not themselves establish that the downloaded artifact meets every delivery specification. Save the inspection result beside the exact archived file.

If dimensions differ from the approved delivery target, determine whether the issue belongs to generation, download handling, transcoding, or a later editing step. Keep these stages separate in the investigation.

Review the display container

A player may fit the file with letterboxing, crop it to fill a box, or stretch it. Inspect the actual integration and final export rather than judging only the source video.

For a fictional horizontal product clip placed inside a vertical card, identify which edges are lost under a fill crop. Check the main subject, hands, captions, and any planned logo area throughout the action. A crop that looks fine in the first frame may fail after the camera moves.

Use the platform’s current layout preview where appropriate. Do not assume that all destinations share one safe area or one overlay arrangement.

Acceptance criteria

A completed review records requested ratio, decoded dimensions, display handling, crop settings, and time-coded subject visibility checks. Approve the final artifact and placement together.

When the crop fails, choose a reviewed alternative such as a different edit, a designed background around the video, or a newly approved generation. Avoid silently stretching the image to force a fit.

Continue with camera movement review, video resolution, and download validation when diagnosing the full delivery path.

Continue with quality and testing

• generation model regression checklist: Maintain a release scorecard with real evidence
• generation prompt revision diff: Review the change as a field-level hypothesis
• AI image color drift test: Control viewing conditions before judging color
• Browse all quality and testing guides