Workflow Introduction
This workflow is based on the LTX video model and the MSR Multi-Reference LoRA. It allows users to directly input character reference images and scene reference images for video generation.
Compared with a single-reference video workflow, this workflow is more suitable for scenes that require multiple subjects or a combination of character and environment references. For example, users can provide one image as the male character, one image as the female character, and another image as the scene background, then generate a video where the characters are integrated into the selected environment.
This makes the workflow useful for character interaction, scene fusion, multi-subject video generation, and cinematic storytelling with stronger reference consistency.
Model Configuration
The LTX 2.3 base model and other required LTX 2.3 models are the same as those used in the other LTX 2.3 workflows.
If you have not downloaded the LTX 2.3 models yet, please follow the instructions here:
LTX 2.3 Model Download and Setup Guide
https://dfcine.com/2549.html
Additional LoRA Model
The following LoRA model needs to be downloaded separately.
Model Name:
LTX-2.3-Licon-MSR-V1.safetensors
Hugging Face Model Page:
LiconStudio/LTX-2.3-Multiple-Subject-Reference
Install Location:
ComfyUI/models/loras
