How to train textual inversion with my own dataset based on someone else's trained text inversion embedding?

[zengziru]

I want to train my own textual inversion embedding about limb contortion based on the textual inversion embedding of Deep Negative v1.x from Civitai. How can I do for it? For the training script of textual inversion from diffusers, it seems like it just can start training from scratch.