Kandinsky Pro T2V SFT 10s model (19B) duration test

It looks like generating a short duration preview gives the general style of output. So, it's possible to generate many 1 second previews and choose what you like.

These are at the same seed and prompt at 30 steps (dancer at 50 steps). The only difference is the duration.

This is the highest quality T2V model in the latest Kandinsky release:

https://huggingface.co/kandinskylab/Kandinsky-5.0-T2V-Pro-sft-10s