docs: document LVFace embedder support

LVFace-B_Glint360K.onnx shares ArcFace's I/O contract (112x112 aligned crop ->
L2-normalised 512-d) and input scaling, so it drops in via --arcface-model.
Note the caveat that galleries and calib caches must be rebuilt with the same
embedder used for analysis.

README.md

@@ -48,9 +48,29 @@ Optional flags:
 The ONNX model weights live in `models/` (tracked via Git LFS):
 - `arcface_w600k_r50.onnx` — primary ArcFace embedder
 - `arcface_w600k_mbf.onnx`, `arcface_r18.onnx` — lighter alternatives
+- `LVFace-B_Glint360K.onnx` — LVFace embedder (ViT backbone, ICCV 2025), drop-in
+  ArcFace-compatible: same `[N,3,112,112]` input and 512-d output (see below)
 - `face_detection_yunet_2023mar.onnx` — YuNet face detector
 - `scrfd_500m_bnkps.onnx` — SCRFD face detector
 
+### LVFace
+
+[LVFace](https://github.com/bytedance/LVFace) is a Vision-Transformer face
+recognition model. The `LVFace-B_Glint360K.onnx` export shares ArcFace's I/O
+contract (112×112 aligned BGR crop → L2-normalised 512-d embedding) and its
+`(x − 127.5)/128` input scaling, so it slots straight into the existing embedder
+— just point `--arcface-model` at it:
+
+```bash
+./build/scene_analyze --arcface-model models/LVFace-B_Glint360K.onnx \
+    --gallery gallery.json --input movie.mp4
+```
+
+> **Important:** embeddings from different recognition models are not
+> interchangeable. A gallery (and its calibration cache) must be built with the
+> **same** embedder used for analysis — rebuild the gallery with
+> `--arcface models/LVFace-B_Glint360K.onnx` before analysing with LVFace.
+
 If they are missing (e.g. LFS not fetched), re-download them with:
 
 ```bash