Fast camera, fast subject
A whip-pan following a skater through a bowl: coherent motion and plausible physics with no cut to hide behind.
2 models received different prompt text — show each
Each model is given the prompt its input contract calls for: a multi-reference model is told which reference is which character, a segmented model gets shot segments. The frame, the references and the requested length are identical throughout. This is the honest version of "the same prompt".
H3 Max, H3 ref2v, Wan 2.7, Seedance 2.0, Veo 3.1, LTX-2
The skateboarder carves across the bowl and up the far wall, catches air at the lip and lands cleanly, while the camera whip-pans to follow him and then pulls back fast. Fast, coherent motion with no cuts. The sound of wheels on concrete, no music.
Kling v3 Pro
[{"prompt":"The skateboarder carves across the bowl and up the far wall, catches air at the lip and lands cleanly, while the camera whip-pans to follow him and then pulls back fast. Fast, coherent motion with no cuts. The sound of wheels on concrete, no music.","duration":"5"}]| H3 Max | H3 ref2v | Kling v3 Pro | Wan 2.7 | Seedance 2.0 | Veo 3.1 | LTX-2 | |
|---|---|---|---|---|---|---|---|
| Lines heard | — | — | — | — | — | — | — |
| Voice | model's own | model's own | model's own | model's own | model's own | model's own | model's own |
| Motion | 12.19 | 12.41 | 7.37 | 5.15 | 15.24 | 10.14 | 6.14 |
| Scene changes | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Loudness / peak | -19.2 / 0.1 | -16.9 / 0.2 | -20.9 / -3.4 | -13.7 / -1.3 | -19.8 / 0 | -28.5 / -4.7 | -16.9 / -2.2 |
| Length | 5.18s | 5.18s | 5.04s | 5.04s | 5.06s | 8.00s | 4.92s |
| Frame | 768×1344 | 768×1344 | 1072×1928 | 720×1280 | 720×1280 | 720×1280 | 576×1024 |
| Wait | 6s | 2:35 | 2:51 | 1:54 | 2:55 | 35s | 40s |
| Attempts | 1 | 1 | 1 | 1 | 1 | 2 | 1 |
| Cost | $0.40 | $0.30 | $0.98 | $0.50 | $0.90 | $3.20 | $0.22 |
The read
The hardest input, and the one where the spread is widest. H3 Max delivered the shot as written: a carve across the bowl, air at the lip, a whip-pan that keeps the skater in frame. Veo and Kling stayed coherent with a following camera and plausible physics; Kling added a lens flare and let the skater get small. Seedance was the most energetic and changed framing so fast it reads like cuts, though none were detected. H3 reference-to-video lost scale — the skater fills the frame in one beat and is a speck in the next. Wan put a tiny skater in an enormous empty bowl, and LTX drifted the camera while the skater receded. Motion numbers here run from 5 to 15; above the near-static threshold they rank nothing.
One person's judgment from the takes above, 2026-09-03. The table is the evidence; this is the opinion.
what whisper heard
- H3 Max: [Sound of a rocket taking off]
- H3 ref2v: [BLANK_AUDIO]
- Kling v3 Pro: (upbeat music)
- Wan 2.7: (car engine revving)
- Seedance 2.0: (loud snoring)