A person whispers the word "apple," barely audible. Then speaks it at normal volume. Then shouts it, loud and strained.
Seedance 2.0ByteDance · proprietary
Kling 3.0 OmniKuaishou · proprietary
Veo 3.1Google · proprietary
LTX-2.3Lightricks · open
Ovi 1.1character.ai · open
JavisDiT++JavisDiT · open
MagiHumanGAIR-NLP · open
Per-prompt rubric
Basic standards
Visual presence
- a person
Event: a person speaks a word
Audio presence
- a person
Sound: a spoken word
Key standards
Visual physical commonsense (V-PC)
- The person says the target word 'apple' three distinct times with visible mouth articulation, with vocal effort visibly escalating from a whisper-like mouth shape through normal speech to a shout.
Audio physical commonsense (A-PC)
- The loudness of the spoken word increases monotonically across the three utterances, from a barely-audible whisper through normal conversational volume to a loud strained shout.
Cross-modal physical commonsense (AV-PC)
- Each spoken 'apple' onset aligns in time with a visible mouth articulation.
- The loudest utterance corresponds to the strongest visible vocal effort.
- The softest utterance corresponds to the gentlest visible vocal effort.