SadTalker Checkpoint Verifier

Quickly check whether all SadTalker model files and checkpoint weights are installed correctly. Detect missing files, inspect exact byte sizes, and get the lightweight verification script.

Scan Your SadTalker Folder or Checkpoints

Select your SadTalker repository folder or individual checkpoints/ files.

Safe: Runs entirely in your browser. No files are uploaded to any server.

Installation Mode Requirements

Filter checklist by your intended inference command

StatusCheckpoint File & Target PathTypeExpected SizeAction
Unchecked
SadTalker_V0.0.2_256.safetensors
checkpoints/SadTalker_V0.0.2_256.safetensors
Core 256px audio-driven generator weights. Mandatory for standard generation.
Required
691.5 MB
Download
Unchecked
mapping_00109-model.pth.tar
checkpoints/mapping_00109-model.pth.tar
ExpNet acoustic-to-expression mapping model for 256px resolution.
Required
148.6 MB
Download
Unchecked
BFM_model_front.mat
checkpoints/BFM_Fitting/BFM_model_front.mat
Basel Face Model 3DMM front mesh basis matrix.
Required
25.8 MB
Download
Unchecked
similarity_Lm3D_all.mat
checkpoints/BFM_Fitting/similarity_Lm3D_all.mat
3DMM landmark coordinate similarity matrix for facial alignment.
Required
3.8 KB
Download
Unchecked
SadTalker_V0.0.2_512.safetensors
checkpoints/SadTalker_V0.0.2_512.safetensors
High-resolution generator. Required only when passing --size 512.
Optional (High-Res 512px)
691.5 MB
Download
Unchecked
mapping_00229-model.pth.tar
checkpoints/mapping_00229-model.pth.tar
ExpNet audio mapping model specifically tuned for 512px generation.
Optional (High-Res 512px)
148.3 MB
Download
Unchecked
GFPGANv1.4.pth
gfpgan/weights/GFPGANv1.4.pth
Face restoration network. Required when using --enhancer gfpgan.
Optional (Face Enhancer)
332.5 MB
Download
Unchecked
alignment_WFLW_4HG.pth
gfpgan/weights/alignment_WFLW_4HG.pth
Facexlib 98-point facial landmark detector for eye and mouth alignment.
Optional (Face Enhancer)
185.2 MB
Download
Unchecked
detection_Resnet50_Final.pth
gfpgan/weights/detection_Resnet50_Final.pth
RetinaFace ResNet50 backbone for initial bounding box face localization.
Optional (Face Enhancer)
104.2 MB
Download
Unchecked
parsing_parsenet.pth
gfpgan/weights/parsing_parsenet.pth
BiSeNet facial segmentation network to mask hair, eyes, skin, and lips.
Optional (Face Enhancer)
81.4 MB
Download

Target Directory Structure

Ensure your files match this exact hierarchy. In particular, BFM_Fitting/ must be a direct child of checkpoints/:

SadTalker/                      <-- Project root (where inference.py resides)
├── checkpoints/
│   ├── SadTalker_V0.0.2_256.safetensors    [691.5 MB] (Mandatory)
│   ├── SadTalker_V0.0.2_512.safetensors    [691.5 MB] (Optional for --size 512)
│   ├── mapping_00109-model.pth.tar         [148.6 MB] (Mandatory)
│   ├── mapping_00229-model.pth.tar         [148.3 MB] (Optional for --size 512)
│   └── BFM_Fitting/                        <-- DIRECT subfolder under checkpoints/
│       ├── BFM_model_front.mat             [25.8 MB]
│       └── similarity_Lm3D_all.mat         [3.8 KB]
│
├── gfpgan/
│   └── weights/                            <-- DIRECT subfolder under gfpgan/
│       ├── GFPGANv1.4.pth                  [332.5 MB]
│       ├── alignment_WFLW_4HG.pth          [185.2 MB]
│       ├── detection_Resnet50_Final.pth    [104.2 MB]
│       └── parsing_parsenet.pth            [81.4 MB]
│
├── inference.py
└── verify_models.py                        <-- Place verifier here

Frequently Asked Questions

Related Guides & References