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 (69)
so great to have the V3!!!
TY!
Woah... have found myself using the set often the past weeks... this is cool news! :) ... although I would sometimes wish for a combined Lora or Neg... as civit only allows for 12 ressources... which I shouldn't be using so much in the first place, but sometimes the vision needs to be helped along the way lol ;) Thank you! :)
I think I'll make a couple of combos, thanks for the suggestion and tip!
Thank you for sharing this embedding, this embedding is the only one out of all the others that will work with SD Forge. I can't get any other embeddings to populate my list.
Anyone try this on ponyRealism? any value? thanks
Am I just slow to understanding what's going on with this update? About to download it but it's still confusing me. Been using the XXX and one of the old normal ones for some time, but they were embeds, that I just kept the basic one on and just added the X's when needed. So it's just a Lora now? And how do I make it swap? Are those meant to be the new triggers, and is that how they are written? A lot of info I feel just got tossed out the window and I feel very lost.
nothing new, this V3 is still an embedding. you can use it with the others
@Zovya Odd, it's only giving me a safetensors download, as far as I'm aware, that's a lora file.
@Battia85 They can be safetensors now too. Most people are making them safetensors these days.
@Zovya Gave it a go, and still used it like a lora before I saw this post. https://civitai.com/posts/5457590
Outstanding Embeddings.
So good. Especially when I Inpainting on hands.
Thank you 😍💘
Somehow I still got censored results, maybe there's problem with my other lora. I tried use v2 and its neg again and they worked perfectly without censoring.
safetensor?
When I migrate from HQ V2 to HQ V3, should I completely discard the negative embedding or keep HQ V2 Neg as my negative and just replace the positive embedding?
That's right, you only need the V3 positive. I did that to help those that generate on site because you can't use the negative onsite. But even if you used the negative, it won't hurt anything.
So v3 isn't negative embedding...
does the same thing, but you use it in the positive. Signs of the times, new types of generators and the phasing out of negative prompts.
I feel that V3 is more compatible with other LORA than V2, thanks for the great works.
Hmm, when I add it, compared to typing the usual score...9,8,7 string, I get more cartoonish, less detailed images. Even if I keep source_cartoon in the negative prompt.
P.S. Nevermind. I found it was the tag "3d" in the negatives. Removing it makes model draw more simplistic and even forget how specific items look.
How to actually install into stable diffusion? Explain if I was a child ;) I apologize I'm new to this stuff
just write in negative promt this: embedding:zPDXL3
@Nalgotica I think I got it thank you 🙏
so I download the negative to the stable diffusion embeddings folder > then enter prompt into negative prompt. Do I need both the embedding file and negative or just negative?
@jjhodges89 no, put the file in the stable diffusion embeddings directory. the V3 version is placed in the positive prompt only. depending on what program you're using. for auto1111 and forge, just put zPDXL3 in the positive prompt
@Zovya Gotcha so if I primarily use the photo realistic version with both a embedding file and negative file download options I would still follow what you said? Thanks for replying and I apologize if I'm not understanding
@jjhodges89 Hi if I understand it correctly, we should download both embeddings if it has, in the same embeddings folder. But when prompting, use non-negative embeddings in the positive prompt and negative embeddings in the negative prompt. Check the showcases and see how people use them.
Anyone know how to get both embeddeding onto the civitai generator?
Just click on the 'CREATE' button
@nyaan_chan_ai There is a positive and negative. It seems to only add the negative.
Does V3 require the NSFW embedding as well?
I'd also like to know this!
noice
Newbie here using Forge UI.
Am I supposed to be using more than one embedding at a time with this? For example, High Quality + XXX? Also do I put the trigger word in positive or negative prompt box?
If I'm correct, you can use more than 1 in your prompt but I only use one at the time. More than 1 damages the image generation in my tests. Also ZPDXL2 and below you have to put the trigger word on the positive and the trigger word with the '-neg' word in the negative. For ZPDXL3 you only need the trigger word on the positive.
@nyaan_chan_ai - helpful comment. But I'm still a bit confused. ALL of these under the overall title Pony PDXL Negative Embeddings. Are you saying that just adding "-neg" to any of them makes them neg embeddings, but otherwise they can be used as positive embeddings? Thanks for the clarification if you have the time.
@UlyssesHeart Of course not. You have to download the file and paste it in the embeddings folder. You can found that folder by going to the location in your pc where you installed Forge. then to make sure the software is using the embedding 'Yes or Yes' simply make sure the positive prompt includes this for zPDXL1 or zDXL2 in parenthesis: (embedding:zPDXL1) or (embedding:zPDXL2) and like this in the negative prompt: (embedding:zPDXL1-neg) or (embedding:zPDXL2-neg).
For zPDXL3 you only need to add it on the positive prompt like this (embedding:zPDXL3). No need to add it on the negative prompt according to Zovya.
It works the same for other embeddings like: (embedding:badhands), (embedding:perfecteyes), etc.
I hope it helps.
How do you use in Comfy UI?
(embedding: PonyNE:1) in genative prompt?
(embedding:zPDXL2:1) in positive
(embedding:zPDXL2-neg:1) in negative
@ZikViM There is only one file, why would you use this in positive if its negative embedding?
@OniChap this applies to versions v1 and v2, for v3 its just a positive,anyway concept how to use embeddings in comfyui ui its like that
Would the embedding go into the negative I assume for the high quality? Sorry for asking a silly question.
"V3 only needs one positive embedding" ??? These are negative embeddings? then it says you can mix and match whatever embeddings you need? But I thought only one is needed? Furthermore it says High Quality V3 removes censoring so why is there is an XXX rating model? Do I need both? and if I want it photo realistic, do I need all 3? Super confused :(
Positive and negative embeddings just mean where they are placed, so positive embeddings are just placed in the positive prompt and negative prompts in the negative prompt.
You only "need" one to get the effect of said embedding, but if you want to you can use several since they don't overlap or just ruin the image.
They have different intended effects, so for example the V3 is from my understandin more general use quality improvement, pg rating makes the images less prone to be NSFW while the NSFW embedding will make the images more prone to be NSFW.
Some models are better or worse at nsfw so they can be good depending on what model you use.
These all work really well, however, the RL positive embedding has a very strong female-bias.
Even putting 1man as prompt, rPDXLrl >0.2 turns the subject female most of the time. IT DOES make her look more realistic, though.
photo real didn't want to work
Can anyone tell me exactly what zPDXL3 OR zPDXL does? Just higher quality right. What about for xxx, how would I type it? I put it in the prompt and wow what a good difference!
i am curious too. it's been a month, have you figured it out yourself?
So how does the XXX one work exactly? I put in both the positive and the negative, and it (should) improve NSFW results? Or do I just put one of them in?
Does it work with Illustrious models?, and if not. Will there be an Illustrious version of this? ;o
Where are all the safetensor versions of these models? It's weird that you provide safetensors for only one
Why does this turn legs into rubber and make them weird?
If I wanted a photorealistic AND xxx image, should I use both POS/NEG Photorealistic PDXL and then both POS/NEG PDXXXL?
What will happen if I do? do I HAVE to use both? Seems a little redundant but I am not sure!
XXX version has two files. Same things, and Just extension diff?
Or, same features, and zPDXLxxx is for positive prompt? This is confusing because the title of this post is pony PDXL 'Negative' Embeddings.
Crazy isn't it? One of the best examples of terrible naming and description I've seen on CivitAI
Not sure what prompt I'm supposed to put this in as some seem to be putting it in neg and others pos. TBH I don't know if it is just my imagination but I do see much better generation but my checkpoint isn't even Pony, but I DO see improvement.
Are these negative embeddings or positives embeddings because the author does not make that clear. And I can't find any of the below examples listing any of these embeddings in either positive or negative prompts. And I thought you didn't have to use the score tags because a lot of the examples continue to use the score tags. It's all very confusing!
download both files to the loras folder, connect as loras, in the positive prompt write the positive word trigger, and in the field with the negative prompt write the negative word trigger. voila. (for example zPDXLxxx, zPDXLrl,)
USAGE (Negative/Positive) - TL;DR
HighQualityV3 - Positive: zPDXL3 Negative: (none)
XXX Rating - Positive: (zPDXL), zPDXLxxx Negative: (zPDXL-neg), zPDXLxxx-neg
PG Rating - Positive: (zPDXL), zPDXLpg Negative: (zPDXL-neg), zPDXLpg-neg
Photo Real - Positive: (zPDXL), zPDXLrl Negative: (zPDXL-neg), zPDXLrl-neg
After looking through the author's sample images and prompts, it appears that both negative and positive triggers are used. Many users have only included one trigger, but I imagine both are required for maximum impact. You could also experiment with negative, positive, or both, along with weights.
To use HQ3, XXX and Photo Real, add each embeddding to the generator and the following to the prompts
Positive: (zPDXL), zPDXL3, zPDXLxxx, zPDXLrl
Negative: (zPDXL-neg), zPDXLxxx-neg, zPDXLrl-neg
excellent explanation sir 😊
So you can mix them up?
@thestockporthatter1883 yes, but try them separately and then altogether to see which combination gives you the best results. also they will have a greater effect on some models than others
I was really confused. But your comment helped me to figure out how these should be used. Thanks and thumb up!
It can't be. Everyone in XXX Rating Gallery is wrong?
Is it an embedding you call with : embedding:PositiveOnly_Pony_zPDXL3,
Or a lora, you put wight 1, and use key word in positive : zPDXL3 ?
Not clear
Only embedding I've used so far, has worked well for me
zPDXLxxx - Positive
zPDXLxxx-neg - Negative
or both negative?
Exactly! As good as this is, the naming and explanation is so incredibly poor
good job tho !
Use the positive embedding names (like (zPDXL), zPDXL3, zPDXLrl, etc.) in your positive prompt.
Use the negative versions (like (zPDXL-neg), zPDXLrl-neg) in your negative prompt to reduce common artifacts.
You can mix several embeddings, or test them separately to find the best combo for your model and style.
Embeddings only work if you have loaded them into your Stable Diffusion setup!
Example:
Positive Prompt: (zPDXL), zPDXL3, pony, beautiful, highly detailed Negative Prompt: (zPDXL-neg), bad anatomy, blurryTry different combinations and weights for best results.
Please upload this to TensorArt, I'm begging you. I cannot replicate my waifu without this.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.