You're welcome! I mainly did them cause I was tired of hunting for words to use on the negative prompt.

You can also try the following format on Automatic1111:

[:embedding:0.5] <- This will make the embedding start later in the rendering process, so you can keep it from altering the original composition. 0.5 for example will mean it will start at 50% of the rendering process. I've found that anything between 0.3 and 0.5 (30-50%) is enough to maintain composition.

If you want it to the embedding to only take place at the start of the rending you can use:

[embedding::0.1] <- This would make the embedding only work at the first 10% of steps, great for composition embedding to keep them from altering final details. For example, an outfit embedding that was trained on japanese women. If you use that format with anything between 0.1 to 0.3 it will often do the outfit and not alter the face.

@Jaxx I don't know if it is capable of that. I've never tried.

@MorganFreeman What do you mean?

The higher the number, more negative tokens they incorporate.

Lately I only use the 4v trained one, because my current model doesn't need a negative prompt that complex.

Hm, they shouldn't create any positives to make that happen. And the last one that was trained, didn't even see people to begin with.

Which one is doing that so I can look into it?

@SnG17 Ahh, gotcha. Yeah, there are negatives for that already.

The name of the TI being used is written above each picture. If it's not being used, it's blank.

They are negative embeddings that you add to your negative prompt. It's for SD1.5 models, you can try using them if your model has trouble making better looking images.

@Malessar No, they just use more negative tokens to hopefully cover more things that can make an image bad. I'd say just grab them all and test to see what is best for the model you're using.