In a warmly lit tea shop, a woman in a green cardigan looks at her friend, sets down a ceramic cup with both hands, and says, "I should have told you sooner." Her friend quietly nods, then reaches across the table and covers her hand. Natural lip-sync, expressive fingers, steady two-shot.

Characters & Dialogue · default settings, one run, 5 s, 720p, 16:9 · Updated September 14, 2026

Not run yet
MiniMax FastH3 V2
P-Video 2 Pro
4.4videomodels.ai score
  • Prompt adherence4.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio4.5

The hand-covering beat and steady two-shot land cleanly, with a slight dialogue variation.

Wan 3.0
3.7videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency3.5
  • Camera & framing—
  • Quality & detail4.0
  • Look & lighting4.0
  • Realism & physics3.5
  • Text rendering—
  • Audio—

Warm setting and two-shot hold up, but the spoken line and hand-off beat are incomplete.

Wan 3.0 Prime
3.8videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.0
  • Realism & physics3.5
  • Text rendering—
  • Audio—

Warm and coherent, but the hand-covering action is reversed.

FastH3 V2
3.8videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The friend never reaches across; the green-cardigan woman covers her hand instead.

Seedance 2.5
4.0videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency4.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The friend touches the shoulder rather than covering her hand.

Seedance 2.0
4.0videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio4.5

The hand cover lands, but the nod is absent and the spoken line is wrong.

Seedance 2.0 Fast
3.7videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio2.5

The comforting contact reads well, but the requested spoken beat and nod are not clear.

Kling 3.0 Turbo
3.3videomodels.ai score
  • Prompt adherence1.5
  • Motion & consistency3.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio2.5

It looks polished, but the requested cup-to-hand-cover action never occurs.

Veo 3.1
4.1videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio4.5

The interaction reads well, but the friend nod is only implied and speech adds words.

Veo 3.1 Fast
3.7videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency4.0
  • Camera & framing3.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio5.0

The apology scene is attractive, but the woman rather than her friend initiates contact.

Happy Horse 1.1
4.1videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The comforting hand cover and exact spoken line land, but the nod is not clear.

MiniMax H3
4.1videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The emotional hand cover and line work, but the required nod is not legible.

Gemini Omni Flash
4.1videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The spoken line and hand cover work, while cup placement and nod remain unclear.

Wan 2.7
4.0videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio5.0

The setting and final contact work, but the reaching role is reversed.

Grok Imagine Video
4.0videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency3.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio5.0

The line and supportive reach land, with visible cup and hand continuity drift.

P-Video 2
4.0videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio5.0

The tender two-shot is strong, but the hand-reaching role is reversed.

Runway Gen-4.5
3.9videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The hand cover arrives, but the cup set-down and nod are not clearly shown.

Hailuo 2.3
3.9videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency4.0
  • Camera & framing5.0
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The cup placement is clear, but the woman rather than her friend reaches first.

Kling 2.6
3.8videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio—

The tea-shop tableau is strong, but the specified apology sequence is incomplete.

FLUX 3
4.2videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

Warm, coherent performance, though the cup placement and nod are understated.

PixVerse V5
4.0videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The final comforting hand cover reads clearly, but the nod and cup set-down are understated.

Seedance 1.5 Pro
4.1videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The audio and final reassuring touch work, but cup placement and nod are not established.

FastH3
4.6videomodels.ai score
  • Prompt adherence4.5
  • Motion & consistency4.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.5
  • Text rendering—
  • Audio5.0

The full emotional sequence and matching spoken line land cleanly; the nod is subtle.

Cosmos 3 Super
3.1videomodels.ai score
  • Prompt adherence1.5
  • Motion & consistency3.0
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics3.0
  • Text rendering—
  • Audio—

The comforting hand cover appears, but the cup set-down, speech, and nod do not.

Lance
2.8videomodels.ai score
  • Prompt adherence1.5
  • Motion & consistency3.0
  • Camera & framing1.5
  • Quality & detail4.5
  • Look & lighting4.0
  • Realism & physics4.0
  • Text rendering—
  • Audio—

A polished solo café shot misses the friend and every required interaction beat.

Wan T2V
3.6videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency3.5
  • Camera & framing4.5
  • Quality & detail4.0
  • Look & lighting4.0
  • Realism & physics3.0
  • Text rendering—
  • Audio—

The hand-cover gesture lands, but garbled subtitles and an unshown nod weaken the sequence.

HunyuanVideo 1.5
3.6videomodels.ai score
  • Prompt adherence2.0
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

A polished two-shot, but the nod and hand-covering response never occur.

Happy Horse 1.0
4.3videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

Convincing spoken line and hand-cover payoff, though the nod and cup set-down remain subtle.

Helios 14B
3.3videomodels.ai score
  • Prompt adherence2.0
  • Motion & consistency3.5
  • Camera & framing4.5
  • Quality & detail3.5
  • Look & lighting4.0
  • Realism & physics3.5
  • Text rendering—
  • Audio—

The two-handed cup placement is readable, but the nod, dialogue, and hand cover are absent.

Seedance 1.0 Pro Fast
4.1videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

Strong visual continuity and hand-cover payoff, but the cup set-down and nod are not fully clear.

Wan 2.6
4.1videomodels.ai score
  • Prompt adherence3.5
  • Motion & consistency4.0
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

A polished, emotionally readable hand-cover gesture, with subtle cup placement and nod.

Kling 3.0 Omni
3.9videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency4.0
  • Camera & framing3.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio5.0

The spoken line is accurate, but the nod is absent and the close insert breaks the steady two-shot.

P-Video
3.6videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency3.5
  • Camera & framing3.5
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio—

The warm hand-contact ending reads, but the nod and exact hand cover remain ambiguous.

Kling 2.5 Turbo
3.7videomodels.ai score
  • Prompt adherence3.0
  • Motion & consistency3.5
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The cup set-down and hand cover read clearly, but the nod and speech are not evidenced.

Seedance 1.0 Pro
3.1videomodels.ai score
  • Prompt adherence2.0
  • Motion & consistency2.5
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.0
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The tea-room two-shot is clean, but the key apology actions do not occur.

PixVerse V4.5
3.5videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency3.5
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.5
  • Realism & physics3.5
  • Text rendering—
  • Audio—

The supportive hand cover lands, but the cup action and nod are unclear.

PixVerse V4
3.7videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency3.5
  • Camera & framing4.5
  • Quality & detail4.5
  • Look & lighting4.5
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The hand-covering moment works, but the two-handed set-down and nod are absent.

Hailuo 02
3.3videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency3.0
  • Camera & framing4.0
  • Quality & detail3.5
  • Look & lighting4.0
  • Realism & physics4.0
  • Text rendering—
  • Audio—

The spoken beat is approximated, but the nod and hand-cover are never clear.

Hailuo 01 Director
3.3videomodels.ai score
  • Prompt adherence2.5
  • Motion & consistency3.0
  • Camera & framing4.0
  • Quality & detail4.0
  • Look & lighting4.0
  • Realism & physics3.5
  • Text rendering—
  • Audio—

The apology mood is present, but the cup stays held and the subtitle is corrupted.