Feedback for F1

I made a dozen videos with F1, and noticed a few clear differences in comparison to the original model:

1. Movement is much more fast and dynamic — which could be a good thing, if it can be toned down with a prompt. But I'm not sure it can be controlled effectively.
2. Skin in particular (and bright moving surfaces in general) often shows a sudden flickering effect, with textures/shadows changing rapidly in a way similar to flickering when they shouldn’t change. It seems to happen in almost all F1 videos. In comparison, with the original model, it rarely happens even with fast motion.
3. The model tends to produce much less coherent bodies, with limbs breaking or disappearing, or going through the body, and with upper and lower body parts moving in different directions. That’s something that doesn’t happen with the original model.
4. F1 seems to be at most 15% faster than the original model on my RTX 4060 Ti 16GB VRAM.

Given all the problems with F1 and the small performance gain, it doesn’t feel like F1 provides much benefit over the original model.
Example:
https://github.com/user-attachments/assets/729b5fe9-df7c-4d43-89b8-b0ef571cd11d

Start by reproducing the linked F1 example and comparing it with the original model on the reported RTX 4060 Ti 16GB setup. Measure the reported motion, flicker, body coherence, and generation-speed differences; done would require a confirmed cause and an agreed change addressing the regressions.

Written by the indexing model from the issue text.