Nodes/ComfyUI-GGUF-Loader/Empty LTX-2.5 AV Latent (Batch) ⚡
ComfyUI Node

Empty LTX-2.5 AV Latent (Batch) ⚡

Joint video+audio latent in LTX-2.5 geometry: video [B,128,(len-1)//8+1,H/32,W/32] and audio [B,z,n_latents,bins] read from the audio VAE. Use this, not the MiniMax H3 empty-AV node, to drive an LTX-2.5 sampler.

By ChrisColeTech·Created 13 days ago·Updated about 4 hours ago· 6
Empty LTX-2.5 AV Latent (Batch) ⚡
  • audio_vae
  • LATENT
width768
height512
length97
frame_rate24.00
batch_size1
Category🤖 CCTech/LTX-2.5

Inputs (6)

NameTypeDefaultDescription
audio_vaeVAEltx-2.5-audio-vae-bf16.safetensors. Only its geometry is read here (latent channels, frequency bins, latents per second) — no encoding happens, so it costs nothing.
widthINT76864–16384
heightINT51264–16384
lengthINT971–16384Frame count. The video VAE compresses 8:1 in time with a causal first frame, so 8k+1 values (9, 97, 121...) tile exactly.
frame_rateFLOAT24.001–120Sets the clip duration, which sets the audio latent length. Use the same value on LTXV Conditioning or the two streams drift apart.
batch_sizeINT11–64Clips per job, each with its own noise from the sampler seed. The LTX-2.5 DiT is batch-aware, so unlike MiniMax H3 no patch node is needed — but VRAM scales with the batch.

Outputs (1)

NameTypeDescription
LATENTLATENT