QuickTalk¶
Support Status¶
| Item | Value |
|---|---|
| Model ID | quicktalk |
| Backend | local |
| Evidence level | Built in, verified |
| Best for | Local realtime adapter, development reference, QuickTalk asset validation |
Recommended Hardware¶
Local CUDA GPU. Validate mock first, then connect QuickTalk assets and a template video.
Weights¶
The QuickTalk local adapter loads weights in the OpenTalking process and does not require OmniRT. Place the model weights, HuBERT files, InsightFace assets, and caches under repository-root models/quicktalk/.
Directory Layout¶
models/
quicktalk/
checkpoints/
quicktalk.pth or 256.onnx
repair.npy
chinese-hubert-large/
pytorch_model.bin
auxiliary/models/buffalo_l/ or auxiliary_min/
det_10g.onnx
If you already have a legacy asset bundle organized as hdModule/checkpoints/, OPENTALKING_QUICKTALK_ASSET_ROOT may point either to the parent directory or to hdModule; the adapter normalizes it to the directory that contains checkpoints/.
Configuration¶
OPENTALKING_QUICKTALK_ASSET_ROOT=/absolute/path/to/opentalking/models/quicktalk
# Optional: built-in QuickTalk avatars declare template_video in their manifests; custom avatars may override it here.
# OPENTALKING_QUICKTALK_TEMPLATE_VIDEO=/absolute/path/to/template.mp4
OPENTALKING_QUICKTALK_WORKER_CACHE=1
OPENTALKING_TORCH_DEVICE=cuda:0
Avatar manifest:
{
"model_type": "quicktalk",
"metadata": {
"asset_root": "/absolute/path/to/opentalking/models/quicktalk",
"template_video": "/absolute/path/to/template.mp4"
}
}
Start¶
export OPENTALKING_TORCH_DEVICE=cuda:0
export OPENTALKING_QUICKTALK_ASSET_ROOT="$DIGITAL_HUMAN_HOME/opentalking/models/quicktalk"
export OPENTALKING_QUICKTALK_WORKER_CACHE=1
cd "$DIGITAL_HUMAN_HOME/opentalking"
bash scripts/start_unified.sh --backend local --model quicktalk --api-port 8210 --web-port 5280
Open http://localhost:5280, select the QuickTalk Local avatar, and select the quicktalk model. If you omit --web-port, the default frontend URL is http://localhost:5173.
Prepare Avatar Cache¶
QuickTalk generates per-avatar runtime cache files:
examples/avatars/<avatar>/quicktalk/template_<width>x<height>.mp4examples/avatars/<avatar>/quicktalk/face_cache_v3_<width>x<height>.npz
These files depend on the local model runtime and avatar contents. Treat them as deployment artifacts and do not commit them to the source repository. To prepare them ahead of time:
cd "$DIGITAL_HUMAN_HOME/opentalking"
opentalking-prepare-cache \
--model quicktalk \
--avatars-root examples/avatars \
--quicktalk-model-root models/quicktalk \
--device cuda:0 \
--model-backend pth \
--verify
To prepare one avatar:
opentalking-prepare-cache \
--model quicktalk \
--avatars-root examples/avatars \
--avatar singer \
--quicktalk-model-root models/quicktalk \
--device cuda:0 \
--model-backend pth \
--verify
/models Verification¶
Expected:
Common Errors¶
| Symptom | Action |
|---|---|
connected=false |
Check QuickTalk dependencies, asset paths, and OPENTALKING_TORCH_DEVICE. |
| Long first turn | Enable OPENTALKING_QUICKTALK_WORKER_CACHE=1. |
| Avatar load failure | asset_root and template_video must be reachable absolute paths. |