This LORA is focused on generating huge head pieces in the style of 16th to 20th century Russian and Slavic countries. I wish I knew more about the art history of these things. I would probably do a more respectful job on the model. Please let me know if I trod upon a sacred or culturally important design element.

Fixed an error in the trigger words: should have been sl4v1ccr0wn3y3s [updated]

Kokoshnik

Thanks to @netgoblin for filling me in about the Kokoshnik or traditional traditional Russian headdress worn by women and girls to accompany the trapezoidal dress called a sarafan.

According to Wikipedia the word kokoshnik describes a great variety of headdresses worn throughout Russia, including the cylindrical hats of Veliky Novgorod, two-pointed nimbus kika of Vladimir, triangular kika of Kostroma, small pearl hats of Kargopol, and scarlet kokoshniks of Moscow.

As you can see from the dataset, I was working with this expansive definition of the Kokoshnik. I also included modern artistic interpretations of the same general theme. To me the whole look is very exotic, sensual and unsettling. Especially with the veils which cover the eyes.

I don't want to get too far down a clinical art history of Russian costume design here. I just wanted a cool look. From what little I have learned so far, however, I think this concept deserves another version that explores the more archetypical elements of the kokoshnik design. Look for that in Version 5.

As it is, the LORA generates some cool images. When I combine it with my face paint LORAs, I get some creepy looks. Works best in my opinion at full strength. I'm including the training set so you can see what I am talking about. This first version was just captioned "sl4v1ccr0wn". In the next version I am going to add more detail to the captioning of the training set so I can capture some of the more interesting features.

V1 All Together

In this first version I put all the dataset images together captioned with the keyword "sl4v1ccr0wn". The dataset images are heterogenous. Although you can get some coherent images that look cool, Some of the most interesting features are buried.

V2 (skipped)

V3 Four keywords

Broke out the looks into four separate keywords

"sl4v1ccr0wn3y3s" which generates images of people wearing the ornate headpieces with veils, cages and/or tassels covering their eyes. Delightfully creepy.

"sl4v1ccr0wnwh1t3" which generates a certain style of headpiece that tends to be jewel-encrusted.

"sl4v1ccr0wn0th3r" which generates images that cleave more closely to the traditional headpieces you see in 20th century images of Russian ethnic dress

"sl4v1ccr0wncr33p" which generates huge ornate crowns that intersect with the wearer. Especially with "man" in the prompt.

You can also mix and match the prompt keywords to get interesting effects.

V4 Trained at Higher Resolution

Everything you love about V3 but trained at a higher resolution. Use V3 or V4 depending on what resolution you are generating images at. In my experience the deliver similar results.

V5 Coming soon!