silkySaido

This is a merge of a curated set of checkpoints to incorporate data that's been trained onto the Pony v6 checkpoint.

The specific merge technique used aims to maintain as many strengths of each merged model while minimizing conflict and "drowning" from tension between them. This is a custom technique I have developed.

In addition to merging in as much high quality trained checkpoint "data" as possible, a second phase of merging is done to achieve a cohesive yet adaptive visual style (heavily inspired by MomoiroPony, WAI-CUTE, and boleromix).

All preview images are generated with the same fixed seed and no post processing or up-scaling is applied.

Prompt

Pony prompt guidelines as per usual, for example:

Positive:

score_9, score_8_up, score_7_up, source_anime

Negative:

score_6, score_5, score_4, source_furry, source_cartoon

Inference

I recommend 20-30 steps, Euler A or DPM++ SDE Karras