LTX-2.3 text connectors
Description
Takes raw stacked per-layer text encoder hidden states and produces the video and audio text embeddings for the DiT: per-token RMS norm, per-modality sqrt(dim ratio) rescaling and projection, then a per-modality 1D connector transformer.
Usage
ltx23_text_connectors(caption_channels = 3840L, text_proj_in_factor = 49L,
video_connector_num_attention_heads = 32L,
video_connector_attention_head_dim = 128L,
video_connector_num_layers = 8L,
video_connector_num_learnable_registers = 128L,
video_gated_attn = TRUE,
audio_connector_num_attention_heads = 32L,
audio_connector_attention_head_dim = 64L,
audio_connector_num_layers = 8L,
audio_connector_num_learnable_registers = 128L,
audio_gated_attn = TRUE,
connector_rope_base_seq_len = 4096L, rope_theta = 10000,
rope_double_precision = TRUE, rope_type = "split",
video_hidden_dim = 4096L, audio_hidden_dim = 2048L,
proj_bias = TRUE)
Arguments
caption_channels: Integer. Text encoder hidden size (3840 for Gemma3-12B).text_proj_in_factor: Integer. Number of stacked hidden states (num_layers + 1 = 49 for Gemma3-12B).video_connector_num_learnable_registers: Integer or NULL.video_gated_attn: Logical.audio_connector_num_learnable_registers: Integer or NULL.audio_gated_attn: Logical.proj_bias: Logical. Projection bias (TRUE for LTX-2.3).video_connector_num_attention_heads,video_connector_attention_head_dim,video_connector_num_layers: Video connector shape (LTX-2.3: 32 x 128, 8 layers).audio_connector_num_attention_heads,audio_connector_attention_head_dim,audio_connector_num_layers: Audio connector shape (LTX-2.3: 32 x 64, 8 layers).connector_rope_base_seq_len,rope_theta,rope_double_precision,rope_type: RoPE config.video_hidden_dim,audio_hidden_dim: Integers. Projection targets (DiT inner dims: 4096 / 2048).