News

LTX-Best-Face-ID: Preserving Identity in LTX-2.3 Video

A specialized LoRA has been released for the LTX-2.3 (22B) video model, providing high accuracy in preserving character appearance.

Compiled by Sergey KostenchukPublished 2026-07-23Updated 2026-07-23

2026-07-23 Multimodal HuggingFace

Expanded analysis for this story

Open the longform version with context, source trail, and what changed.

Read longform

Alissonerdx/LTX-Best-Face-ID · Hugging Face Source Video file

🎨 The new LTX-Best-Face-ID LoRA for the LTX-2.3 (22B) video model allows for character identity preservation.

The technology utilizes the Reference-to-Video (ref_t2v) method and the TASS-RoPE mechanism to transfer facial features using ArcFace identity loss. Modes such as Face ID Base for close-ups and Character-Sheet for preserving clothing and physique are available.

🌍 It solves a key problem in generative video — character consistency.

👤 You can now create videos featuring a specific person simply by uploading their photo.

Source 1: https://huggingface.co/Alissonerdx/LTX-Best-Face-ID

Sources