test_actor

Character References Rig Voice Cloner Training 3D Assets

Train a Video LoRA and generate face expressions for animation

1. Training Data

Dataset Mix

uploaded

Gemini

Unknown

Real uploads should own the core front, profile, and back slots. Gemini is best as filler, cleanup, and motion coverage.

Best Next Captures

Coverage is full. If identity still drifts, replace the weakest Gemini-generated back or profile pose with a cleaner upload before retraining.

Training Signal

Every pose can be generated. Uploads are still the best option for the core body views.

Use uploads for the core body views. Autofill is best for secondary motion poses or fast placeholder coverage, not your strongest identity anchors.

2. Train LoRA

Target:

Wan2.2 and Hybrid-WAN use this LoRA mostly to hold body identity, costume silhouette, and rear/profile consistency. Stage 2 motion helps faces during generation, but it does not replace strong body coverage.

Trained LoRAs

Expressions can also be generated directly. Upload only when you want to override a weak result.

Recordings

Delete LoRA

Delete character?

This will permanently delete test_actor and all associated files including references, recordings, animation LoRA data, and performances.

Share to Project