Struct VideoVae 

pub struct VideoVae { /* private fields */ }

Implementations

impl VideoVae

pub fn from_cmf(model: &Arc<CmfModel>) -> Result<Self, String>

pub fn decode( &self, z: &[f32], t_lat: usize, zh: usize, zw: usize, ) -> (Vec<f32>, usize)

Normalized latents [z_channels, t_lat, zh, zw] → RGB in [0, 1], [3, frames, zh·16, zw·16].

pub fn decode_image(&self, z: &[f32], zh: usize, zw: usize) -> Vec<f32>

MiniMax-H3-Image: one normalized latent frame [z_channels, zh, zw] → RGB in [0, 1], [3, zh·16, zw·16]. The decoder takes the latent as it is (no clip of copies around it), tiled the way the video one is.

pub fn spatial_ratio(&self) -> usize

pub fn temporal_ratio(&self) -> usize