When @alexds9 released the long awaited Babes 3.0, I grabbed a copy and started playing to see how it would fit in and work with one or more of my own checkpoint merges. I figured it would at least merge into Kitsch-In-Sync pretty well since that's my "Mutt" model anyway. But, I was also hopeful that it would fit into the rest since, after all, Babes 1.1 and 2.0 have been a part of most of my merges from the start.

After a few hours of tinkering and merging in some of my sanity and concept bases, I was sitting on this model - ready to then be merged into something else. But THIS model was too good to waste by hiding it inside something else - so I decided to release it on its own.

XenoBABES

This checkpoint has a very similar look and feel as Babes 3.0 but, with the various elements I've added, it should (maybe):

Generate more "finished" looking pieces with simpler prompts. Complex prompts actually turn out very similar - and in comparisons, it sort of went back and forth between which won. (I'm posting a few comparisons in the example pix). But on simpler prompts - XenoBABES is tuned to hopefully fill in the blanks better without the need to get as descriptive to get things going.
It works well on lower end machines and at smaller starting sizes (I've been using 384x512 with a basic latent 1.5x upscale on my No GPU rig for testing - and they work great.)
It takes directions well - and understands a few concepts better. Things like "looking at cell phone" worked 3 out of 4 times on this, while I couldn't get her to take her hand away from her ear as if talking on the phone on BABES.
It takes controlnet poses and LoRas really well - images in the gallery will pop up shortly. I need some time to put them together and since I like to test on "Worst Case" scenarios - that can take a while. But initial tests on my GPU system all passed with flying colors.

Babes or XenoBABES?

If you like more prompting and not having it add things you didn't ask for - BABES 3.0 is probably a better choice. If you like easier prompting and letting the AI fill in the blanks (hopefully smartly and pleasantly) then XenoBABES might be something to try. And, of course, if you're on a lower end machine - this model is probably going to work a little better for you.

Prompting

All of the various trained tags I tried from the list on the Babes 3.0 page all seemed to work - except some of the "Might be Improved" people weren't that great (on either). It makes pretty people, but they don't necessarily have enough images and tagging to call them great.

If you're familiar with my other Xeno brand checkpoints - prompting is similar. Who, what, where - and it should start making nice things. Add some general styles or names of artists, photogs, directors, and it goes all sorts of fun places.

This does NOT know all of the things most of my other models know - yet. I wanted a version of this release while it was clean and still had all of Alex's training up at the top level. But it does know a lot of them. Experiment. And have fun.

NOTE: This model, as it stands now, doesn't go to many of the "darker" places that my other models can take you. Creepy Slime and Monsters tend to be playful creatures, That sort of thing.

Like it? Buy me a beer.

I'm using Civitai's trainer - so Buzz Point tips are helpful too - even just a few add up.

Want Tips, Tricks, Pix of Things We Can't Show on Civitai?
Join the XenoEngine Discord to discuss tips and tricks, complimentary Lora's and Embeds, and to post (and see) images not allowed here on Civitai.