Fix model

[mykeehu]

@l8doku patch. Read more here:
#21 (comment)

[l8doku]

I didn't submit my patch as a pull request because it's a workaround, not a real fix. It assumes that all models use "openai/clip-vit-large-patch14" as their text encoder. But, for example, Stable Diffusion 2.0+ uses OpenCLIP instead.

I don't know how to fix this properly, though. This is not to say that this pull request shouldn't be merged, more to state clearly what it does.

[mykeehu]

I put it in PR because two weeks ago this solution helped several people, but @AUTOMATIC1111 hasn't looked at it yet, until then this extension is available, it couldn't be used without it. If Automatic has a better solution, I will withdraw the PR.