Why nobody cared about BitDance?

I remember that "BitDance is an autoregressive multimodal generative model" there are two versions, one with 16 visual tokens that work in parallel and another with 64 per step, in theory,thid should make the model accurate than any current model, the preview examples on their page looked interesting, but there's no official on Comfyui, there are some custom nodes but only to use it with bf16 and with 16gb vram is not working at all (bleeding to cpu making it super slow). I could only test it on a huggingface space and of.