V1.0 - Initial release.
V1.5 - Improves on the previous formula and reduces tokens used from 5 to 2.
Character renditions coming up with skin that's just a tad too perfect? Use this negative embedding to dial in more skin texture and character.
Only uses 5 tokens so as not to flood your negative token space.
Images were generated on my upcoming model.
Description
V1.5 - Improves on the previous formula and reduces tokens used from 5 to 2.
I'm considering this as an intermediate step since I'm not quite happy with the resulting contrast, but it does prevent character renders looking like addicts.
Character renditions coming up with skin that's just a tad too perfect? Use this negative embedding to dial in more skin texture and character.
Only uses 2 tokens so as not to flood your negative token space.
Images were generated on my upcoming model.
FAQ
Comments (5)
Where do you put the trigger word?
Positive prompts or negative prompts?
It's a negative embedding, so the negative prompt.
In the context of embeddings, if it says "neg" or "negative" anywhere then you're safe to assume it goes into the negative prompt. For instance in the title of this embed model it's "SkinPerfection_Neg" letting you know that it goes in the negative. In the description it says "Use this negative embedding to dial in more skin texture and character.", and "Only uses 5 tokens so as not to flood your negative token space."
Hope that helps.
When i'm not sure if an embed model is a positive or negative I do these steps:
① Double check the model page to see if there is any mention of it in the title, description, trigger word, or the "About this version" dropdown that I missed.
② Look through the images in the gallery of the model to see their generation parameters. Civitai only supports automatically grabbing generation data from specific generation tools, meaning sometimes its blank.
③ Look through the discussion section, both the initial questions AND the replies (the little 💬 button on the discussion) to see if any other user has the info I need.
④ Just try it out in the negative and then in the positive, using the same generation parameters I can compare the results to see which looks better.
safetensor?
Hugging Face has an embedding converter online,
@straycabin228
"~too perfect?"
When I first saw this sentence, I misunderstood it to mean "perfect realism."
In fact, it was "perfect in the opposite sense."
If it had been "~too smooth?" or "~too retouch?" I don't think I would have misunderstood.
Details
Files
SkinPerfection_NegV15.pt
Mirrors
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
skinperfectionNegv15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
SkinPerfection_NegV15.pt
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.
