What the scene asks for
A living room, two friends on a couch, already engaged with each other; one continuous medium-wide shot.
- At about three seconds, one of them produces exactly one brief fart.
- Both react briefly and naturally.
- They return to what they were doing before the clip ends.
- No second emission, no intelligible speech, no music.
A summary of the scene definition every configuration was given. The exact prompt each model family received is published with the results; the human reviewers do not see it either.
The three factual questions
Every reviewer answers these in this order, yes, no or unclear, before rating anything. Unclear is a real answer and is never counted as a no.
-
Was exactly one fart-like event clearly present? event question
Primary — asks whether the event happened; never N/A
-
Did the two people briefly react to it?
Dependent — may become N/A when its prerequisite is absent
- N/A — no event to react to — when question 1 is “no” (the scene’s event is absent)
- N/A — no event/reaction relationship can be judged — at any time (the relevant behaviour cannot be seen or heard)
-
Did they naturally recover and return toward their prior interaction, without another emission? primary causal question
Dependent — may become N/A when its prerequisite is absent
- N/A — no event to recover from — when question 1 is “no” (the scene’s event is absent)
Then three ratings, the same in every scene
- How real did the fart sound?
1 Obviously fake · 2 Pretty fake · 3 Believable but imperfect · 4 Very believable · 5 Completely real - Did the timing make sense? — This means whether the sound/action/reaction happened in a believable order at the right moment.
1 Way off · 2 Mostly off · 3 Mostly right · 4 Very good · 5 Exactly right - Did the people and movement feel natural?
1 Very unnatural · 2 Mostly unnatural · 3 Mostly natural · 4 Very natural · 5 Completely natural