mstar.model.cosmos3.components

mstar.model.cosmos3.components#

Cosmos3 backbone components.

Modules

conditioning

Conditioning-frame preprocessing for image-to-video.

packing

Joint-sequence packing for Cosmos3 generation (ported from the diffusers Cosmos3OmniPipeline).

reasoner

Prompt-side plumbing for the Cosmos3-Edge reasoner: media preprocessing, placeholder expansion and the 3D mRoPE positions of a VLM prompt.

sound_tokenizer

Cosmos3 AVAE sound tokenizer (decoder-only), ported from the diffusers-format sound_tokenizer/ checkpoint component that ships with the Cosmos3 models.

vision

The Cosmos3-Edge reasoner's vision tower and projector.