Modelref2va turbo LoRA / TaoMate ref2va turbo
TaoMate ref2va
An audio-visual joint guidance LoRA for video generation that extends the reference-to-video (ref2v) paradigm by incorporating audio signals as an additional conditioning input. A turbo variant reduces inference to three steps, making it suitable for audio-driven reference-image-to-video generation tasks.
Timeline (last 90 days)
Sep 16
In a comparison of three LoRAs, TaoMate ref2va uses 3 steps with approximately 4–5 minutes per clip and the best timing and volume consistency; larryvrh turbo uses 4/5 steps and approximately 5 minutes; lightx2v ref2v uses 4 steps and approximately 6 minutes.
Unverified50%