Pony Diffusion XL Complete Beginner’s Guide (NSFW Focus)

Optimal settings for Pony Diffusion XL:

Settings I use

• Weight: 1.0
  • However, do not hesitate to try different weights. 🙂 If I have a different weight recommendation for the model, you can see it in the model version features.
• Checkpoint: ponyDiffusionV6XL_v6StartWithThisOne or other Pony Variations
  • You can use this LoRa on Civitai only with Pony based models. However, you can download the model and use it with any model you want (of course it must be XL).
• Resolution: 832×1216 / 1216×832 / 1024×1024
  • These are the dimensions in the features of the civitai image generator. Since XL models are trained with 1024×1024 images, you can get the best results if the total of the measurements you provide is 2048 pixels. (832 + 1216 = 2048)
• Sampling Steps: 26+
  • I usually use 30.
• CFG Scale: 7.5
  • I usually use 7.5. If you reduce this, you will pay less attention to the prompts and you may get the results you do not want. On the contrary, if you increase it, you may still get bad printouts as it will give more importance to the prompts. Don’t hesitate to try.
• Upcscale: Use what you love.
  • I use, 4x-UltraSharp, 4x_foolhardy_Remacri, RealESRGAN_x4plus_anime_6B. (Stable Diffusion WebUI only, BetterWaifu does not support upscaling yet)
• Sampling Methods: Use what you love.
  • I use, Eular-a, DPM++ 2M SDE, DDIM

Triggers: You can see the triggers in the area with model information. Generally, I add at least one trigger to each of my models (whether necessary or not). Additionally, do not forget to review the images I have produced as examples.

NOT: We collect all kinds of images (Cosplay, 3D, CGI, Real, Realistic, Semi-Realistic, Anime, Draw, Sketch etc.) for a character. Therefore, as the prompts show similarity, the outputs tend to shift to the image type of the relevant image. For example, as the prompt you wrote resembles the prompt of the image I used in the training, the output shifts to the style of the similar image. To prevent this, write the style you do not want to see as negative. For example, if you do not want a “cgi” style output, add a “cgi” prompt to negative.

---

Some Resource

First of all, before continuing the article, I will give some links and explain what they are.

• PonyXL Notes
  • Provides detailed information about PonyXL
• Pony Diffusion XL v6 Innate Character Lists
  • An excel file prepared by @marusame. Be sure to check it out.
• Pony Cheatsheet v2
  • @BrutalPixels A cheatsheet article by @BrutalPixels. You’ll definitely want to take a look.

Some Recommendations Prompts That Can Increase Your Production Quality

Quality tags for Pony v6 and variations: (Use in Prompt section)

score_9, score_8_up, score_7_up, score_6_up, derpibooru_p_95, 

You may want to write the score_6_up prompt as negative. I use it this way.

Extra Quality Prompts: (Use in Prompt section)

detailed eyes, beautiful, detailed background, perfect eyes, 

Use it if you want anime style: (Use in Prompt section)

source_anime, very aesthetic, anime screencap, anime coloring, 

Use it if you want realistic style: (Use in Prompt section)

photography, realistic sun light and shadows, photorealism, UHD --ar 9:16 --style raw --stylize 700, 

rating_safe, cinematic, cinematic photo, close-up, portrait, orange rim lighting, atmospheric, bokeh, dynamic angle, vibrant lighting, dramatic shadows, highly detailed, realistic, high contrast, extremely detailed, photorealistic, professional, detailed skin, depth of field, film grain,

Use it if you want realistic style: (Use in Negative section)

(score_3_up, score_4_up, score_5_up), sketch, drawing, cartoon, greyscale, monochrome, anime, 3d, cgi, source_cartoon, source_furry, source_pony, source_anime

Try the prompts until you get the output you want. Examining images produced by other people is also a good way to learn.

These below will draw or rather (filter) your prompt and image to these specific datasets, OR you can put these in the negative.

source_pony
source_furry
source_anime
source_cartoon
rating_safe
rating_questionable
rating_explicit

Score prompts, you can use them all, or actually use a selection of what group of the scored dataset to use, less means a tighter dataset but could yield better or worse results.

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up

score_9, score_8_up, score_7_up, score_6_up

score_9,

source, rating, and score will constrain the images used to those sources or ratings, and may help in drawing out a character, such as if you know its from an anime, or cartoon.

you can think of the score as a part of the dataset that just include images with that rating up to 9, ie. if you choose a lower score you will only pull from those lower rated iamges. You can also put some of the “score’s” in the negative! an example is

score_6, score_5, score_4, chromatic abberation, artifacts, ugly, bad image,

For different angles:

from_below, 
from_above,
from_side,
close-up, 
upper body, 
portrait, 
contrapposto

General Negative prompt for Pony: (Use in Negative section)

score_6, score_5, score_4, derpibooru_p_low, lowres, bad anatomy, bad hands, text, cropped, worst quality, low quality, jpeg artifacts, signature, watermark, username, blurry, artist name, mosaic censoring, twitter username, logo, limited palette, monochrome, (Blurred poorly drawn eyes), (out of frame), (bad quality eyes), (asymmetric eyes), long neck, elongated neck, blurred eyes, undetailed eyes, ugly face, blurred, grainy, cut off, oversharpened, sign, amateur, extra limbs, dull colours, boring, lacklustre, bad art, text, abominations, more than 2 legs, more than 2 hand, fused hands, bad proportions,colorless, glitch, bad face, distorted face, messed up eyes, deformed, extra limb, extra finger, bad hands, broken finger, black and white,

Or you can Negative Embeddeds.

If you have any suggestions to improve this article, feel free to write.

Enjoy!

Source

• Pony Diffusion V6 XL Prompting Resources and Info