Sign In

GDP_PonyXL

183
2.2k
88
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
481
Reviews
Published
Aug 25, 2024
Base Model
Pony
Usage Tips
Clip Skip: 2
Hash
AutoV2
3AC1AFEAA2
default creator card background decoration
joe78

v1.1 Update:

I've tried to bring some more realism back into my original model by merging a few more realistic models in at varying weights. Seems kinda counterproductive, given the original goal was going for more of an illustration/comic look, but I do what I want. This version responds well to realistic, photorealistic, detailed more effectively than my original, but still maintains a less realistic look by default. That said,

I think I still prefer v1.0.

Here's an example of same seed, same prompt images, version is noted in the prompt:

Example post comparing the two.


Example

Example2

Check the prompts for examples, and have fun trying it out!

The additional models added to the mix included these fantastic merges:

Real Dream which is an amazing model all by itself, make sure to check it out!

Zonkey - Realism provides some great realism, as well as being a bit...horny. Check it out!

See below for the other awesome models included in the mix.

Description

I enjoy making images that are more illustrated/comic style, but more realistic than standard Pony.

Very realistic Pony models seem to need more prompting for the style of image I'm trying to create, and I don't really want the look of OG Pony with ten Loras on top.

So, I took three absolutely amazing Pony models and merged them (which was way easier than I'd thought it was, seriously, I learned the basics in like 10 minutes on YouTube).

This is a merge of three amazing PonyXL models:

The primary model is an excellent realistic model, Godiva. This model produces fantastic images, please check it out!

Godiva was then merged with the awesome Dr. Pibb PonyXL model. I love this checkpoint!

The resultant checkpoint was then merged with another fantastic model, Pony 🦄FaeTality.

I highly recommend consulting the descriptions on the three above models to get the best results.

Parameters

Sampler: Euler A

Steps: 20 - 30

CFG: 5 - 7

VAE: sdxl_vae

Clip Skip: 2

Prompting

I noticed that Godiva especially leans towards depicting younger subjects than I'm interested in. I always include in the Negative Prompt:

child, kid, childish, childlike, loli, 

I do include the score_ tags in the positive prompt. An example prompt:

score_9, score_8_up, score_7_up, (score_9:1.1), (score_8:1.1),
from side,
a woman in a tight black and white striped short sweater dress, glasses, 
arched back, sexy, light brown hair, dark brown eyes, athletic body, narrow waist, sleeveless, depth of field, classroom,
knee up, leaning forward,

Negative prompt:
score_6, score_5, score_4, child, kid, childish, childlike, loli, extra arms, extra limbs, extra hands, muted colors, (monochrome:1.1), (greyscale:1.1), (black and white:1.1), 3d, censored, signature, watermark, username,