Cute Pin-up Style

PetraVoice Inspired

So, I jumped over to Illustrious from Pony, and this is my first successful Lora.

It was heavily inspired by the softer PetraVoice Styles I've seen, but with a cute pin-up style mixed in (most likely from another AI model that I couldn't find the source for).

Together they made something cute, but still similar to the PetraVoice inspiration.

V1 overall has a softer look, and is closer to the PetraVoice inspiration even at lower strengths.

V2 had more of the pin-up style images in the dataset, so it has a sharper style at lower strengths.

Quick Note: In V2 I noticed that some of the skin tone tags, particularly the darker skin tones, tend to create a two-tone skin (especially in the face). This seems to easily be fixed by adding 'two-tone skin' to the negative. I can only assume it's something from Illustrious v1.0 that is interfering with the lora, as V1 was trained on Illustrious v0.1 and didn't seem to have this issue (at least not as apparent).

Although I tagged for the pin-up style images, using the 'pin-up' tag alone will get you a mixed bag, so using 'full body' does a better job. At strength 1.0 you will still get a soft style, but it has more influence from the other (sharper) training images. While at higher strengths (I tested up to 1.5), you get something closer to the classic PetraVoice style.

v1 with side-by-side strength 1.0 and strength 1.5

v2 with side-by-side strength 1.0 and strength 1.5

You can also use the 'portrait tag' to specifically get close-up portrait images, again the higher strength gives something closer to the PetraVoice style.

v1 with side-by-side strength 1.0 and strength 1.5

v2 with side-by-side strength 1.0 and strength 1.5

One final note, this model may lean more towards producing characters wearing revealing clothing, underwear, and nudity. So prompting for specific clothing, or using NSFW or nude in negative prompts may help avoid this. WAI Illustrious v13 especially seems to produce more NSFW/Nude images unless a specific clothing type is prompted, so I personally use WAI Illustrious v12.

V1 was trained using Illustrious-XL v0.1 while V2 used Illustrious-XL v1.0, but all generations are made using WAI-NSFW-illustrious-SDXL v12.

All of my models may be used for your own training and/or merges.