better quality by training

[p3l4h0]

will there be a possibility to insert several pictures of the same person to train it? sometimes you won't get such great results as with deepfacelab. and if it can use multiple images (and multiple created iterations), you might get similar good results as with deepfacelab or apples animoji

I'm not sure how the training in FOOM is made https://github.com/AliaksandrSiarohin/first-order-model#training

[alievk]

It is possible, if you have a dataset with such pictures. A big dataset.

[p3l4h0]

does it come with it, or do you happen to have instructions for it? for single videod i found out how it works with deepfacelab. but live with avatarify not yet.

thanks

[tonyj-th]

I am also very interested to learn how to create and apply the dataset to use with avatarify

[alievk]

does it come with it, or do you happen to have instructions for it? for single videod i found out how it works with deepfacelab. but live with avatarify not yet.

thanks

No, it doesn't and I don't have instructions for it. But I would start from crafting such a dataset from Voxceleb.

[Zeephremia]

Any updates?

[mtae]

Is there a plug and play way to provide more than a single image as training to avatarify? (For instance when you are trying to swap faces with a friend and have an abundance of training data of both the source and the target)

[JohanAR]

Is there a plug and play way to provide more than a single image as training to avatarify?

Just to clarify something, Avatarify doesn't do any training at all. The photo you provide is used as input for a pre-trained network. To train your own network you would have to use other generic AI tools, and it's a very computationally expensive process so it would take a very long time without very expensive special purpose hardware :)