I highly recommend that you try all the versions to find the one that best suits what you are looking for, there are no better versions than others, personally I use all of them.

Pros and Cons of each version:

V1:

Pros: Better asses, shadings and bodies.

Cons: The inside of the ass is usually darker if used at a very high intensity, I recommend not going above 1.

V2:

Pros: Probably good at generating sexual scenes.

Cons: Style can sometimes break for no apparent reason.

V2 Stable:

Pros: As the name suggests, it is more stable than the conventional V2, it is epoch 8/20 (V2 is epoch 20, but maybe it broke due to overtraining), also can generate good sexual scenes.

Cons: The dark part of the ass still appears, but now it leans towards a pinkish color, I recommend generating at intensity 0.8 or less to avoid that problem, also the anuses can have a weird shape sometimes (but it should happen less if used at intensity 0.8).

Rebirth:

Pros: Created specifically to stylize V1's style, trained only and exclusively with images generated with V1 (no human art used), it is good at what V1 is good at and without the previously mentioned problem.

Cons: It may be less flexible because it was trained using less variety of images, most of which were intended to create better bodies.

V3 It's too new to be able to give an opinion on it yet.

It is important to know that the images presented were generated using the "Haruka v2" model, exclusive of PixAi (From V1-V2 to Rebirth).

The V3 images were generated using WAI-NSFW.

Here's the official model link on PixAi if you wanna use it there:

V1:

https://pixai.art/model/1887663145934382379-GoonerMix

V2:

https://pixai.art/model/1893643179833296622-GoonerMix-v2 (Epochs 8-20).

Rebirth:

https://pixai.art/model/1899980243013910003-GoonerMix-Rebirth

V3:

https://pixai.art/model/1915447645072896754-GoonerMix-V3 (Epochs 9-30).

They were also generated with 17-25 steps and Face Fix.

I recommend using a negative prompt that includes 'text', 'speech bubbles', 'patreon username', 'artist name' and 'watermark' (for V2).

Training data info:

V1:

100 images.

(All clean).

Trained on: PixAI using Illustrious v1.0.

V2 (all epochs):

250 images.

(Most of them not clean).

Trained on: CivitAI using Illustrious v0.1.

Rebirth:

100 images generated using the V1 (no human art used), most of them created to stylize the model and make better bodies.

Trained on: PixAI using Illustrious v2.0.

V3:

207 images.

(Cleaned an cropped to 1:1).

Trained on: CivitAi using WAI-NSFW v14.

This LoRA is a mix of styles created from a dataset who has illustrations from some well-known NSFW artists.

I use the illustrations of: Astroodragon, Blackwhiplash, D-Art, Fellatrix and Geulimykun (or Ykun), all credits to them.

Note: If something explained here is not well understood, you can ask for details in the comments, it is probably because I cannot explain myself or because English is not my native language (I apologize for that).