Which AI video model do you actually prefer?

I keep preferring a clip because I already like the name on it. Kling, Wan, Seedance — the logo arrives before the motion does. A blind comparison is the only way I trust my own eyes, and even then this is four takes of one person, not a league table.

Four complete takes of the same man, glasses, and bottle shelves. They are unlabeled. Lengths differ (2.750 s, 3.208 s, 3.583 s, 3.733 s). They are comparable generations, not locked-off duplicates.

In a previous personal blind quiz, Kling 4.0 was my pick. That score is mine, from a different test. It does not seed the tally on this page. Visitor votes start at zero.

Vote or skip · Method

The four takes

Six pairs, one choice each. Order is shuffled per visitor. You do not have to finish — skip and see identities plus whatever public counts exist, with no vote attached to you.

Cost and time from one log

Separate from the vote. These numbers are from my 6 November 2026 log, not a live price crawl. I am not mixing any edit captions into this table.

Model$/sTimeQuiz
Gemini Omni Flash 1.2$0.12363 sNo
Kling 4.0$0.235322 sYes
MiniMax H3 Max$0.11211 sYes
Seedance 2.5$1.164251 sYes
Wan 3.0 Prime$0.280114 sYes

From my test log, last updated 2026-11-06. Not verified current pricing, and not a universal benchmark. Per-second rate is not the total bill or credits spent. Generation time is that job’s wall clock, not billed seconds. Voice is omitted. Gemini Omni Flash 1.2 is in the log and not in these four takes. Quiz means the unlabeled files on this page.

How this is counted

One anonymous cookie, one stored choice per pair, hashed left/right order. Replays do not double-count. Public totals come from the server when storage is connected — not from localStorage dressed up as a crowd. Sample size is disclosed. Four takes, six pairs; this does not sort five model families. Gemini Omni Flash 1.2 is in the cost log and not in these four takes. There is no author flag on the URL.

Identities were matched to a labeled comparison edit by duration, motion, and audio — not by guessing look. The quiz files themselves have no names on them. Prompt and whether the jobs were matched are still unknown.

Not in this quiz

A second clip is a Well-branded office story: different face, intro and outro cards, no vendor model name. I am not guessing the generator.

A separate ~29.3 s clip with Well intro and outro, a different person, and no vendor model name on screen. I am not assigning it to Gemini Omni Flash 1.2 or to Well-as-a-model. It is not in the quiz.