First off, let me say sorry if this is a mess I aren't great at making tutorials and long form media, but I noticed some people trying to get into Pony model and not fully understanding it, so I wanted to show some ways to help improve your images. Also, since this is an AI group I won't explain basic AI stuff as I hope you understand that I normally do Loli work (I'm playing Eri so of course :P), but for this I will use Adult Eri as examples just for people who aren't into them. Lastly I will be using Forge UI so sorry if you use Comfy style. With that said, I hope I do a decent job anyway, let's start!
Chapter 1: Pony Model
Pony is a fine-tuned XL model to work with both SFW and NSFW work images. It can make Humans, Ferals and Furry as well and not just that if you use it right you can make it look realistic like a photo, cartoony, anime or even 3D.
An Example image (Without Inpainting)
Pony Diffusion V6 XL and it's VAE are here.
However, I like to use AutismMix Confetti SDXL, so I recommend trying both out and seeing which you like.
Whichever one you use, you still need this VAE
Chapter 2: Prompts for Pony
One of the confusing thing about using Pony is prompts unlike SD1.5 or other SDXL Models, Pony uses Scores/Source/Ratings.
Let's start with Scores
Scores
in their prompts Pony is trained by images that are manually reviewed by people who give the images from score 9 to score 0 the higher, the better. (At least what I've been told, so don't take this as 100% fact, do research :) ).
So let's do an example of an image without any scores in prompts. (Again no Inpainting.)
Well, it does look okay, but it can be better.
So what do I do to fix this? Scores!
Scores will look like this score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, score_4, score_3. score_2, score_1, score_0.
"Wait a second." You are saying. "Why are two versions of score 4" You notice. Well to tell you the truth I don't know lol, but my guess is Score_4_up is like these are good enough to work while score_4 and lower isn't the best.
So what you want to do is this, put your scores in prompt, but up score_9 to score_4_up.
It should look like this:
But with the other score? You put them in negatives, silly ^^ so by the end it should look.
But of course the is more work than just scores, so for this I will give my basic prompts and negatives :)
- Prompts: score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, hi-res, extremely detailed,best quality, masterpiece, uncensored, detailed background, amazing background, detailed shading, detailed lighting, 8K wallpaper, detailed face, detailed body
- Negatives: score_1, score_2, score_3, score_4, low-res, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer disgits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers: 1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, 3D, Render, 3D Render, Blush,
A note for negatives: 3D,Render, 3D Render is there as I don't want to do 3D looking images nothing wrong with them just a choice, feel free to remove them. Blush is there as I noticed Pony has a habit to make everything blush which isn't an issue if you want that, but annoying when you don't so again feel free to remove it, but I recommend putting it there if you don't want blush.
Okay so if you're using the prompts and negatives I suggested you should have this.
(Ignore the Lora that's just so I can get Eri images)
Now in prompts you put what you want so I'm just doing a basic adult Eri's for this Tutorial after I put what I want I have this at the end.
So you may notice I went down and done something called BREAK, this isn't a pony thing this just something I like to do separate sections, I heard it's meant to help AI within steps, but I just like to do it have it cleaned is all.
So the let me see the finished product! (Not using Inpaint)
That looks much better and more cleaned up (Background is a bit of a mess, but I did just put city as a quick example)
See that's why scores are important, of course feel free to experiment with it :)
So next are Source and Ratings, I'll do them together as they're quick.
Source & Ratings
So pretty much Source is where the AI will focus the images, for example I use Source_Anime this will help find better Anime like pictures, but if you're looking for Furry stuff you likely want Source_Furry they are four types of sources.
- Source_Anime
- Source_Furry
- Source_Pony
- Source_Cartoon
Source_Pony is MLP stuff, as that's what Pony started out as, of course. You can use different sources still, so cartoon character can still use source_anime. So it should look like this.
Of course, if you're using something like source_furry you still gotta let the AI know you're making a furry, so best to put furry in prompts.
Lastly is rating like source this helps the AI sort through images, but by how lewd they are, only three of them to go for.
- rating_safe
- rating_questionable
- rating_explicit
It should be quite obvious what those do, it should look like this.
I'm using safe as I want to make no lewd images right now, of course the AI might still try to, but that's AI for ya.
Chapter 3: Settings
Now that we're done with prompts for Pony, Next should be settings.
Pony is quite simple with settings really, for example I use these.
I recommend not using Hi-res Fix as it causes issues IMO if they are anything need fixing we will use inpainting which I will show soon. Pony has low steps, normally around 24, but I find 30 a good place. HeightxWidth you can do 1024x1024, but I like the 1212x832 you can also put it as 832x1212 if you want more of a portrait style. Euler a is the method I was told best for pony, but you're free to try other stuff, the same with all settings these are set in stone and worse case just doesn't work. CFG I recommend around 4 to 8 but your choice.
That's really it for settings.
Chapter 4: Inpainting & IMG to IMG
You noticed I've mentioned images on this Tutorial that I have not used Inpainting, if you understand what that is well you don't need to be at this point so you can skip if you want or just read to see how I do it. If you're using comfyUI or online tools I can't help 100% with it as I'm using Forge UI (Automatic1111 users should be able to follow along, but I recommend switching to Forge UI).
Inpainting is where you mask a certain point of the image and can be used to try to fix errors or give the image a better look. Let's do an AI gen with no inpaint as our example.
Looks good already even without it, but what if they are something's you don't like about it, for me, I think I can do the eyes better so you take it inpainting.
Under the preview of the image, they are icons you want to click the circled one
This will take you to Inpainting, but first settings of Inpainting,
Make sure you set up to look like this, but HeightxWidth to what you made your image in Text to Img. Denoising Strength will change how much of the image you masked, which I will get into soon, but make sure everything is correct.
So I want to edit my eyes so how you notice the image that is there you can mask stuff on it and zoom in, so I'll zoom in to the image and mask the eyes.
So it should look like this, then you just click generate again.
There you go if you compare the images you notice the eyes look more lively now, but now the eye is over the hair, thats a bit strange, I could just keep generating to get what I like, but let's move it to Img to Img. Best way to do that is to left-click and drag the preview image and go over the left image and let go, this should change the image to your new image (Make sure to do this if you want to do more changes, but happy with one change. They are likely a better inpaiting tutorial then this one.) Then, once you've done that, click to Img to Img
I recommend turning down the Denoising strength to something smaller, I like to start around 0.5 then click generate
And there we go it fixed the eye to go behind the hair, (even given her bathing suit a logo), but what I like to-do is going back to inpainting with the new image keep the denoise low and go over what I did again, but this is optional. That's it for Inpainting, sorry if it isn't great sorry if it isn't super helpful.
Bonus Chapter: Loras & Embeddings Recommendations.
I'm taking you know how Loras & Embeddings work and where to need to be put, so I won't go over that sorry, but if you do need help just let me know :) This part is just to recommend a few Lora's & Embeddings
I recommend you get them from Civitai.com and make sure you use Pony Loras and Embeddings you can try to use SD1.5 and SDXL ones, but I doubt they will work graet.
Pony PDXL Negative Embeddings: I recommend High Quality V3 (I didn't even know it updated lol) and XXX can use them same time, or seprate.
Age Slider LoRA | PonyXL SDXL: Good if you want Loli or Shota or if you want to look older works better with Autismmix, but you can get it working with the default Pony Model.
Styles for Pony Diffusion V6 XL (Not Artists styles): Great if you're trying to get a certain style.
My Hero Academia: I'm a huge MHA fan, so of course I'll shill for it :P
Eri (壊理) - My Hero Academia (僕のヒーローアカデミア): It's me ^^ of course I'm going to recommend it.
Closing Thoughts
Sorry if this isn't helpful, I did my best, but I do hope I got some people to understand or at least able to figure out some stuff themselves. Have a good day :)