I've been using first-order-motion-model to animate artbreeder characters, but it has a few issues (such as not transferring separate eyebrow movements [this can be solved by retraining the model with more keypoints, as I now understand: fomm current vox checkpoint uses 10], target face being distorted when the source head is sideways). Unfortunately, I don't have enough VRAM and I'm not sure how to run Face Reenactment in Google Colab, but if someone wants to perform a test and is able to run Face Reenactment, I could prepare some results from the first-order-motion-model and send a link here to a target image, source video, and the result, so we could compare.
I've been using
first-order-motion-model
to animate artbreeder characters, but it has a few issues (such as not transferring separate eyebrow movements [this can be solved by retraining the model with more keypoints, as I now understand: fomm current vox checkpoint uses 10], target face being distorted when the source head is sideways). Unfortunately, I don't have enough VRAM and I'm not sure how to runFace Reenactment
in Google Colab, but if someone wants to perform a test and is able to runFace Reenactment
, I could prepare some results from thefirst-order-motion-model
and send a link here to a target image, source video, and the result, so we could compare.