Three video models tested on running physics, Gemini looked best
Seedance 2.0, , and were tested with the same running video request. The scene asked for a side-tracking shot of a sprinter, which makes mistakes in body motion, weight, and clothing easy to notice. handled the request cleanly, avoided wrong safety blocks, and produced the most believable body movement.
Its weakness was a slightly slow look, as if the video had fewer frames than expected. Seedance 2.0 made the most polished image, with stronger lighting and a more cinematic look, but its movement was not the most accurate. was harder to use because it wrongly blocked a normal running scene as adult content and sometimes misunderstood the request.
When it did generate, the lighting and frame rate were strong, but the body movement looked unstable. The comparison was easy because all three models worked through one , so only the model name had to change.
Key points
- gave the most believable running motion.
- Seedance 2.0 produced the best-looking image but not the best physics.
- had false safety blocks and weaker prompt understanding.
- One made the side-by-side test much easier.
- Video model choice still depends on whether visual style or motion accuracy matters more.