Every tool gets the same input and the same scoring rubric, so differences come from the model and not from our choices. Until a tool’s clip test is published, its profile shows facts from the vendor’s own pages only, and no score.
1. Vendor facts
For each tool we read its pricing page, terms of service, refund policy and content or safety policy. We record video modes, plan prices, how credits work, payment methods, the billing descriptor if stated, refund terms, the rule on real-person likeness, and the age-check method. Each fact links to the page it came from, with the date we read it.
When a vendor does not publish something, we say “not published” instead of guessing. When a page could only be read from its source code or a search result, the profile says so.
2. The clip test
- Start frame. One non-explicit image of a fictional adult, made in Prompt Lab. The same frame goes into every tool that accepts an image. Tools that only animate their own generated images get the closest equivalent made inside the tool, and the profile says so.
- Prompt. One director prompt from Prompt Lab, changed only where a tool needs its own syntax. We publish the exact prompt with each test.
- Length. A 5-second clip, or the nearest length the tool allows. Longer clips are tested separately and labelled.
- Plan. The cheapest plan that unlocks video, bought at list price.
3. Scoring rubric
| Measure | What we look at | Scale |
|---|---|---|
| Face consistency | Does the last frame show the same person as frame 1: face shape, eyes, hairline, skin texture. | 0 to 10. 10 means no visible drift. |
| Motion | Does the clip follow the requested camera move, with natural movement and no jitter. | 0 to 10 |
| Artifacts | Warped hands or limbs, melting textures, flicker, background morphing. | 0 to 10. 10 means none seen. |
| Refusals | Ten prompts that step from everyday scenes to explicit adult content, always with fictional adults. We count blocks, blurs and silent rewrites. | Count out of 10 |
| Price per clip | Plan price divided by the 5-second clips its monthly credits buy, or the cheapest credit pack if that costs less. | USD |
Two passes are scored separately and averaged. Where the two passes differ by more than two points, we run a third. Any score in a profile links back to the published test.
4. What we never test
- Uploading photos of real people, including ourselves, to tools that allow it.
- Any prompt that suggests a person under 18.
- Ways to get around a tool’s own safety rules.
5. Re-checks
We re-read vendor pages at least every 60 days and whenever a reader or vendor reports a change. The date of the last check is shown at the top of each profile. Corrections follow our editorial policy.