PonyRAP is a flexible Pony model merge made from merging Pony models with a little few Illustrious models and then fine tuning with a custom image set (Please see below for some of the fine-tuned tags). I can produce realistic people, animals, furry, landscapes. The model is primary merged and trained to produce realistic images and keep as many of the danbooru tags as possible; it can do fantasy and realistic style as prompted.

Click "Show More" for recommended settings.

Click "Show More" for list of the major tag trained in this release.

Note: Showcase uses Hires. Of the three version, choose this version when Hires. Fix cannot be used.

Recommended Settings:

Steps: 40
CFG scale: 6
Samplers: DPM++ 3M SDE, Euler a
Hires. fix
- Upscaler: 4x_NMKD-Siax_200k or 4x-UltraSharp
- Steps: 20
- Denoise strength: 0.25-30

Positive Prompt

score_9, score_8_up, score_7_up, {prompt}, photograph \(medium\), ultra-detailed, (film grain), high quality

Negative Prompt

score_6, score_5, score_4, artist name, censored, masaic censoring, patreaon username, watermark, bar censor

Trained Tags (most prominent tags)

Older women tags: milf, mature woman, indoors, old woman, cellulite, stretch marks

Other feminine tags: lingerie (with a significant number of common tags to identify associated undergarments)

Animal tags: gorilla, male gorilla, female gorilla (recommend using the tag "black fur" though not trained)
General Concepts:

older woman (milf, mature woman, etc..),
lingerie (standard undergarments and sexy clothing tags)
clothed woman (better clothes)
gorilla (cartoons and photographs)

Future Training Goals:

My plan is to continue to train human, animal, and landscape concepts and test to make sure that I don't lose the flexibility of the model through careful selections of the tags. This will be the last version of PonyRAP and I'll release the finetuned model as a new release series.

Version 2.0 and 1.0 (Merges)

PonyRAP is a flexible Pony model merge made from combining many Pony models, realistic furry, anime, and illustrious. The model is primary merged to produce realistic images and keep as many of the danbooru tags as possible; it can do fantasy and realistic-anime style as prompted. PonyRAP also does a good job of reproducing character LoRAs.

Showcase uses Hires. Fix

Recommended Settings:

Steps: 30-40
CFG scale: 6
Samplers: DPM++ 3M SDE, Euler a
Hires. fix
- Upscaler: 4x_NMKD-Siax_200k or 4x-UltraSharp
- Steps: 20
- Denoise strength: 0.25-30

Positive Prompt

photo, score_9, score_8_up, score_7_up, {prompt}, photo background, source_photo, [highly|ultra] detailed, (analog film grain), high quality

Negative Prompt

score_6, score_5, score_4, low_resolution, low_quality, blurry, blur, illustration, painting, sketch, cartoon, anime