A model trained on 3,000+ PonyXL images - these were tagged manually with descriptions and danbooru tags.

I've only uploaded fp8 and nf4 because those are the only models I can make on my measly 4080S <3

This was created by merging https://civarchive.com/models/891281/ponyish-flux-sfwnsfw into Flux Dev fp8.
The intention in future releases to is to merge additional LoRAs to avoid "retraining" on large datasets.
Currently it takes me 5 days to train the equivalent LoRA, so it's time to make the jump to a checkpoint.

Recommended Settings:
CFG: 2.5
Steps: 20-30