Textual Inversion - from images (png/webp)

[creeduk]

Any way to use these or anyway to convert back to a pt file? I have a few including one of these early ones Victorian lace.

[WASasquatch]

Textual Inversions are "embeddings". Put them in your /ComfyUI/models/embdeddings/ folder.

.bin embdeddings from Diffusers can be renamed to .pt in WebUI without issues, I'm not sure about ComfyUI though.

As far as training in ComfyUI, not yet, though would be cool to be able to train TIs and Hypernetworks, and LORAs.

[fractal-fumbler]

i think point of question is that to use embeddings that are in png formats
for example https://huggingface.co/Conflictx/Chempunk/blob/main/ChemPunk-96.png

as i understood atm ComfyUI didn't recognize such as embeddings

[WASasquatch]

i think point of question is that to use embeddings that are in png formats for example https://huggingface.co/Conflictx/Chempunk/blob/main/ChemPunk-96.png

as i understood atm ComfyUI didn't recognize such as embeddings

Isn't that just PNGInfo prompt from WebUI? Not an embedding. That specifically is sample image from the training process, isn't it?

[fractal-fumbler]

no, it's not preview.
it's embedding. it can be used as embedding in a1111 webui.

[WASasquatch]

no, it's not preview. it's embedding. it can be used as embedding in a1111 webui.

That's interesting, how? You just put it in embeddings folder?

[fractal-fumbler]

yes, same as pt or safetensor file