I’m sharing this technical experiment with you, which has pushed me to the limits of what Seedance 2 can achieve today. The pieces we generate with AI tend to have a polished look and very aesthetic, distinctive movements—a finish that sometimes feels too “clean” or artificial. For this sequence, I wanted to try something completely different. I set out to create a raw, rough piece, without unnecessary frills or artifice. The main challenge was the workflow: everything was generated using pure Text-to-Video, without using any reference images or supporting keyframes. To achieve this, I had to forego automation and focus on absolute control over the prompting. Using Seedance 2, without music and without external visual aids, I had to explain to the machine—with an extreme level of technical detail—every impact, every camera movement, the weight of the armor, and the fatigue on the faces. What you see is the result of that effort to tame the model toward a realistic and visceral aesthetic, far removed from what we normally see in generative AI.