ichioda/LTX-2.3-VBVR-LoRa-ComfyUI

VBVR-LTX2.3 (ComfyUI)

ComfyUI-compatible build of the Video-Reason/VBVR-LTX2.3-diffsynth LoRA, repackaged from the original DiffSynth (PEFT) format into the key layout expected by ComfyUI's LTX LoRA loader.

All credit goes to the original authors — see the upstream repository and the VBVR project.

Installation

1. Copy vbvr_ltx23_comfyui.safetensors into ComfyUI/models/loras/
2. Load it with LoraLoader (or LoraLoaderModelOnly) on top of the LTX-2.3 base model

Requirements

• Base model: Lightricks/LTX-2.3
• Node pack: ComfyUI-LTXVideo with LTX-2 support

LTX-2 introduces audio-video joint attention (audio_attn*, audio_to_video_attn, video_to_audio_attn). If your ComfyUI build does not expose these layers, the LoRA will load but those blocks will be no-ops.

Details

• 2304 tensors, bf16, rank 32
• 428 MB
• Weights are identical to the upstream LoRA — only key names were rewritten

License

Inherits the license of the original LoRA. See the upstream repository.