aero_leap 6 hours ago

HunyuanVideo currently only supports text to video, and I wanted to do image to video like luma dream machine. I added image conditioning using a new sampling method and a LoRA with pretty exciting results. This method can also do things like video infilling, and potentially frame infilling to create high frame rate versions of existing videos. Excited to continue research

  • verdverm 6 hours ago

    Safe to assume you have seen their generative 3d mesh model? Anything exciting planned for that? Can we generate 3d videos for something like VR devices?

    • aero_leap 5 hours ago

      I haven't had a chance to try it yet but it looks super cool, definitely excited to play around with it and see what's possible.