Full-pipeline ternary quantized model trained on CANN.
Create short AI videos from text or first and last frames, with synchronized dialogue, effects, and ambience.