This checkpoint was intended as a 'test', its origins lie in Cyberrealistic Pony V6.0, Halcyon Pony V11 and Realistic PonyPhoto V1.0

The mixing ratio is exactly one third and VAE is included.

The surprisingly positive result is (if the ADetailer settings are adhered to) very realistic faces and with perfect weighting I have not had any problems with hands. There are still a few things to consider:

1. with over 125 tokens, it can happen that weighted prompts of up to 1.5 force the VAE to change the clip, which leads to chaotic results. In principle, the checkpoint could also tolerate DPM++SDE, but you will definitely get the best (realistic) results with Euler a!

2. In the NSFW area, the checkpoint starts to react 'bitchy'! The fewer tokens are used in the prompt field, the better the results! It also responds well to Booru tags. But it really comes into its own in the SFW area! You can and should use clear prompts to achieve the desired result, e.g. 'light-fx', 'particles' or 'cheek2cheek', prompts from real life. All three original checkpoints have a good starting point of trained data sets, although Cyberrealistic has given up a bit too much of its pale skin. Please look at the prompts on the example images and note the settings for ADetailer on one of the images!