No need to use score this, and score that. Just use the embeddings instead.
New V3 versions only need the one positive embedding!.
A set of quality enhancing embeddings for Pony SDXL, and other Pony-Adjacent models. You can mix and match any of the embeddings that you feel you need. They don't duplicate each other, so feel free to use as many together as needed.
High Quality V3 (zPDXL3): gives higher quality results and removes annoying censoring.
XXX Rating: tells the Pony model of your choice to allow NSFW content.
PG Rating: tells the Pony model of your choice to try and remove NSFW content for safe images.
Photo Real: tells the Pony model of your choice to use realistic or near realistic renders instead of cartoon or anime.
High Quality V2 (old): designed to give higher quality results and remove censoring.
Doesn't work with non-Pony models. Check your model.
Recommend using both positive and negative embeddings together at strength 1.0 to 2.0. If you can't use both, the Positive version should be take priority.
Description
FAQ
Comments (18)
Do we just put this in the Embeddings folder of A1111? I thought XL couldn't use embeddings tbh.
How was this embedding trained? A bunch of score_x outputs trained as the more "normal" adjectives?
scores and other quality boosters, it's doesn't touch anything concerning styles or rating. those will come shortly
Problem with that though is isn't that going to result in the training set driving the quality tokens unless they're exhaustively captioned?
@i860 ok
Sorry for the dumb question, but what the heck is a pony model?
pony diffusion.... the most annoying model you have to read on every 2nd upload
@stapfschuh hehe, yeah I hear ya
an alternately trained version of SDXL that uses a slightly different way of prompting
@Zovya i do the same different way of prompting but without the need of a special checkpoint. my way of prompting is working on any checkpoint and any version of sd, heck, even outside of sd. check my articles if you want to know more.
The Pony model is the most expressive 2D model, with easy control of the picture and a surprising output
You sure that having a "-" in the neg name is a good idea? SD might read this as the positive with "-neg" added to it. I renamed my version but didn't experiment
If it's not a good idea, someone should test it, then tell the tons of negative TIs that use it. It's the most standard naming we have for negative embeddings. It would be unfortunate if none of them worked.
as per the preview images, it works just great
Zovya and others have been using the "-neg" suffix on negative embeddings forever. It's completely fine.
I've been renaming all my models (checkpoints, LoRAs, and embeddings) to include all kinds of characters, including hyphens, for a very long time. Embeddings will even work with spaces in them, at least on A1111.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.