Kitsch-In-Sync (...and a bag of chips, by Xenos)

Kitsch: something of tawdry design, appearance, or content created to appeal to popular or undiscriminating taste.
Sync: making things work together.

See "About this Version" tab on the right for update notes.

NOTE: With Google cracking down even harder on Stable Diffusion within Colabs, affordable access for us poor hobbyist types has gotten even more difficult. This model was both merged on, tested on, and remixed on my No GPU Huggingface Space. Many models work well on there - while many don't do so well. My most recent checkpoints all work as well as I've been able to make them.

Examples of K-i-S No GPU outputs can be seen here. These have NO upscaling, ADetailer, or any post work done to them. Upscaling (can do up to around 1600px on the long side if you're doing one or two images at a time), ADetailer (included in my space), and standard post work will improve the results even more.

Honestly, on this model, I've found that some of the No GPU renders are actually more pleasing than some of the normal Colab/T4 type renders I normally do.

Like it? Buy me a beer.
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.

Versatile and fun mix of my other models plus some trained checkpoint merges laid in on top. This pushes fairly hard toward American Cartoon/Comic styles, but will take other styles very well - be it from LoRa's or other models, or be it from its own knowledge of art styles, popular culture, and so on.

Because this is a merge of lots of different training styles and methods, complicated pose scenes can end up with some lame hands and even the occasional extra limb. The standard negative prompts and embeddings will help, there. I've also been experimenting with the hand models in ADetailer - but the jury is still out on whether that helps or not. Sometimes it does great and other times it seems to make things worse.

You may also find that some tokens need some (emphasis:1.2) of some sort while others may need to be backed down a bit. This is typically for tokens or concepts with the same trigger that were merged together making them essentially 2x as strong as they probably should be. Something like (de-emphasis:0.5) would bring that token in at half power.

In the end, though, this model is a lot of fun to play with. It often goes in interesting directions when given the freedom to do so, but will also stay on point when desired. There are lots of things to improve (and I'm working on those) but this is definitely worth playing with and seeing what sort of fun things you can make or how it might interpret some of your favorite prompts differently from other models.