"This is excellent. I was very displeased with my Dreambooth results on 2.1. I tried some of the positive embeddings which helped, but this is amazing. Like night and day. Thank you!" - PiyarSquare, Reddit
Have you guys noticed something... odd about the faces in 2.X? Proportions seem a bit off, tiny eyes mashed close together, faces seem stretched and contorted, smiles just look.... bad? I like to call that 2.0 ugh face, and it's a very common problem in 2.0 for some reason. Well, never fear, through a lot of testing, I've come up with a negative embedding that fixes those goofy WTF faces! Just add my new high quality facelift embedding to your negative prompt and wipe those goofy grins away!
Warning - This is a Negative Embedding and is designed to be used in the negative prompt. you're gonna get some seriously wacky results if you use this in your regular prompt!
I am using my AnalogFilm768 embedding for the samples on this page. As with all of my high quality embeddings, this Facelift embedding is highly versatile and flexible, allowing you to craft the scenes you want with little alteration from my embeddings. Check my profile to see all of my other high quality embeddings, and follow me if you want to be notified when I release new content.
💖💖I hope you enjoy using these embeddings in your art as much as I enjoy creating them for the Stable Diffusion community. 🙏🏼🙏🏼If you do like them, can you pretty please click the heart at the top and leave a review? Many thanks! 💖💖
Versions
Neg_Facelift768 - Invoke this embedding from your negative prompt, watch the magic happen. It fixes faces and makes them look more realistic. That's it. It can't fix an already ugly picture, but it sure can turn those SD2.0 clown frowns up-side-down!
---------
A note on the creation of this embedding
This embedding was created through targeted prompt crafting, tokenization and careful merging. No training involved! Because of this method of creation, It is highly versatile and flexible, and should work in a multitude of situations and scenarios. Due to the nature of how its created, it is on the larger side at 19 vectors. This shouldn't cause any problems in AUTO1111 UI, however it may slow down your It/s a bit.
--------
To use these embeddings in AUTOMATIC1111, copy them to your embeddings folder in your stable-diffusion-webui directory. No need to restart, just invoke the embedding from your NEGATIVE prompt. Make sure you are using a compatible model or you will get errors!
--------
Description
Invoke this embedding from your negative prompt, watch the magic happen. It fixes faces and makes them look more realistic. That's it. It can't fix an already ugly picture, but it sure can turn those SD2.0 clown frowns up-side-down!
FAQ
Comments (11)
If I rename the file to (2.0)-Neg_Facelift768.pt can I still invoke it with token call Neg_Facelift768, or does it need to be called with an exact match to the updated filename? Thanks!
Just like with any embedding, you invoke by the name of the file itself. You can rename it whatever you like and that will be the new trigger word, it will still work fine as long as it's matched with a compatible model (2.x).
@socalguitarist does this also work for 2.1?
@aimer Yes, it's not a training so it should work for any 2.0/2.1 models both 512 and 768, and even should work with trained 2.X models, though YMMV on those.
@socalguitarist ok, thanks 👌
Now make a fix yo ugly hands version XD
2.1 isn't too bad with hands honestly, and Facelift does help with hands actually, same concept as with faces, ugly hands are swept under the rug... but SD still doesn't quite know how many fingers humans have, so it's always gonna struggle at least a little bit there.
Some of those preview images are scary as hell.
SD doesn't want to load this for me, not sure how to correct it
This version is for 2.0/2.1, are you using it with the correct model version?
lol one of the faces was so ugly it was hidden as 13+
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.