Fast Negative Embedding

I have heard that negative LoRAs doesn't work in theory. LoRA edits existing weights and embedding adds new word.

Wow, that's too bad. I thought LoRA would take the negative prompt to the next level, but it doesn't seem to work. Hmm.

@Lykon Would you be willing to share a portion of the data set you used? I am trying to make my own negative embedding for a few very specific problems and don't really have a good reference for what my data set should look like

I'm told that training a lora and then adding it to positive prompt with negative weight can successfully "do the opposite" of what it was trained though. Is that true?

@jkroker Yes, there is many "slider LoRAs" currently.

It's not a LoRa, it's an embedding. You should just put the keyword in the negative prompt

@Sentence Thats what it comes up as, I literally copied and pasted it.

Should I just write FastNegativeV2 in Negative Prompt?

@nutronic I think you placed it in the LoRa folder, it's not the correct placement. You should instead place it in the embedding folder and then yes, paste FastNegativeV2 in negative prompt.

@Sentence If I use a pony model instead of sd1.5 will this still work?

Hah, this is one of those head smack moments for me. Of course! Thanks!

I was really struggeling with result of this embedding, thanks dude/dudette!

https://civitai.com/models/1361164/fastnegativesdxl

Exyloz/safeConvert on HG.

@rlbyv2iel187 Found them; sincerest thanks.

All I need now are the safetesnors for: BadDream + UnrealisticDream, Deep Neagtive V1.75, Bad Artist Negative.

@silvdra00563 
I guess, is best if we learn how to convert .pt to .safetensor in a safe way.
I don't try yet, but in Exyloz/safeConvert mention a conversion script witch runs on google colab.
You can find various script.
Also doing the conversion inside offline and restricted VM(plus sandbox?) sounds ok to me. Cheer!

@rlbyv2iel187 It's going to be a while before I try looking again; I tried at least 3 different times in less than a five days, (and because I'm such a genuine amateur) all I got for my trouble was tempted-to-cuss-someone-out levels of frustration.