Removed 131 dead tensors that audio-only inference never touches:
- model.diffusion_model.video_embeddings_connector.* (~3.2 GB, 8 transformer blocks at 4096-dim)
- text_embedding_projection.video_aggregate_embed.* (~1.5 GB)
Runtime audio_only=True path drops these before .to(device) and replaces video_aggregate_embed with a zero-returning dummy. Local A/B inference reproduces identical denoise/decode timing and output shape.
6.71 GB -> 1.94 GB.
Audio-only inference paths replace this Linear with a zero-returning
dummy after load (see ltx_pipelines.utils.blocks.PromptEncoder
with audio_only=True), so the weights are dead on disk. Removing
them brings audio-components from 6.71 GB -> 5.17 GB; behavior
verified bit-identical via local A/B inference.
Revert license section to a short pointer; LICENSE file carries full terms
