---
title: "LTX-Best-Face-ID"
emoji: 🎬
colorFrom: purple
colorTo: pink
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: "Distilled LTX-2.3 identity video from a reference photo"
python_version: "3.12"
startup_duration_timeout: 1h
---

# LTX-Best-Face-ID

Identity-preserving video generation from a reference photo using **LTX-2** with the **LTX-Best-Face-ID** LoRA.

Upload a close-up frontal photo of a person, describe the scene, and get a short video that maintains their identity across frames.

## How it works

This demo uses the [Alissonerdx/LTX-Best-Face-ID](https://huggingface.co/Alissonerdx/LTX-Best-Face-ID) LoRA on top of the [LTX-2](https://huggingface.co/Lightricks/LTX-2) base model. Identity is injected through **overlap reference conditioning** with **TASS-RoPE** (source-phase RoPE), replicating the ComfyUI BFSNodes logic in a diffusers-style pipeline.

## Tips for best results

- Use a **clean, frontal, well-lit close-up** as the reference
- Let the **Prompt Enhancer** add identity attributes automatically
- Prompts are automatically prefixed with `ref_t2v:` for the model