Is there a plan to open source the training code?

[Sallieimt9n4x2odn1Nolie]

[wangxiang1230]

Hi, thanks for your interest in UniAnimate. Due to the company's open source restrictions, our code can only release the inference code, you can try to try to use v-prediction to train our model on your own data, and we very much welcome your peers to update their models to our repository.

[ak01user]

sorry,what is the v-prediction mean,could you give us some more things about training/finetuning?

[wangxiang1230]

sorry,what is the v-prediction mean,could you give us some more things about training/finetuning?

Hi, v-prediction means velocity prediction rather than noise prediction (refer to https://medium.com/@zljdanceholic/three-stable-diffusion-training-losses-x0-epsilon-and-v-prediction-126de920eb73). Our model is trained on 16/32 frames with a video resolution of 512x768.