AI VIDEO MODEL
中文
Editorial standards

How we distinguish specifications, advice and actual tests

We publish documented specifications and clearly identified workflow advice. Unless a page includes an explicit test protocol and results, it is not a hands-on review or a quality benchmark.

Evidence and dates

Capability claims use official model cards, help documentation, release announcements or provider information pages. A release announcement proves what the provider announced at that time; it does not prove that every feature is enabled for every account. Marketing claims of superiority are not repeated as independent findings.

We record the substantive review date and link the source. If a source describes a feature as planned, preview-only or coming soon, that qualification remains visible. Where access cannot be confirmed, we say so.

How comparison fields are normalized

  • Duration means one generation unless the field explicitly says editing, extension or assembled sequence.
  • Generation resolution is kept separate from upscaling and export dimensions.
  • A platform-level tool does not automatically establish a capability of every model in that platform.
  • API access and web access are separate fields.
  • Credits, subscriptions and API dollars are not converted without a disclosed rate.
  • Undisclosed data stays unconfirmed; it is not filled from a similarly named model.

What our recommendations mean

Workflow suggestions are editorial inferences from documented controls and the needs described in the guide. Prompt examples, shot plans and sample budgets are original illustrative material. They are not generated-output evidence or measured performance results.

If we later publish hands-on tests, each result must include the tested model and access surface, inputs, settings, attempt count, review rubric, charge and representative outputs. Small pilots will be identified as small pilots.

Updates and corrections

We update a page when the relevant evidence or practical guidance changes. Dates are not advanced merely to look fresh. Readers can report a specific correction through the public repository linked on the About page. Provider status and prices should be checked again before a production commitment.

The machine-readable content files are generated from the same source as the visible pages. They do not contain hidden rankings or unsupported claims intended only for AI systems.