🗣️ edge-lipsync trainer
web wrapper around
tools/make_avatar.sh
Avatar name *
Talking video *
— or pick an existing video —
Upload the talking-head video used for training.
How many idle videos?
(0 = reuse the talking video as the idle loop)
Advanced training options
Max steps
Batch size
Learning rate
GPU id
Init checkpoint (optional)
Upload WebGPU bundle to S3 (
--upload
)
Force retrain even if checkpoint exists
Skip face-detection pre-check
Clean up all local files after build (
--cleanup
, needs upload)
tools/make_avatar.sh …
â–¶ Start training
Jobs
↻
idle
â– Stop
Waiting to start…
step 0 / 0
train_loss
–
val_recon
–
val_mouth
–
val_temporal
–