Queen Tyr'ahnee - Character
Very angy and very sexy. It took a lot of retrains + testing to get this one to work correctly. The character has no mouth, a very specific mark on her face, a lot of details around her eyes AND most importantly, most of the pictures are in cartoony style.
Don't expect it to replicate the character too well on realistic checkpoints. Works perfectly fine on anime and cartoony ones.
Use it at weight 1 or 0.9.
Trigger word: queentyr
Tips:
Always accompany with: black skin, colored skin, facial mark
For her classic look: white hair, angry, purple eyes
For her outfit: circlet, bracer, neck ring, yellow bra, gold belt, pelvic curtain
For her feet: purple footwear
For models where her mouth is being rendered: Positive: no mouth | Negative: lips, lipstick (Add weights if necessary)
Her hair is highly customizable, her clothing is customizable to a degree, eyes are way too hard to change.
Most important - Her eyes will look fucked up to some degree. You will need to upscale to fix it. Maybe even inpaint in some checkpoints.
Description
FAQ
Comments (11)
Who is this character?
That Queen obsessed with Duck Dogers, old cartoon, would recommend but ignore the ending.
An old crush reappears, thanks for the queen, and I won't accept the ending ever
I remember that show barely aired when I was a kid :( But I absolutely love her design.
Used to love this show
Can you tell more about your training method? Because several days ago I tried to train lora for Valentine from Skullgirls game, and lora refused to generate her pupil correctly (she has a pupil that looks like a cross)
Honestly I just follow this guide, and always use LoRa https://civitai.com/models/22530/guide-make-your-own-loras-easy-and-free. One thing I can recommend to your for specific characteristics such as that, is to remember removing that from the captions. For example, if your tag should be "symbol-shaped pupils, cross-shaped pupils" or something like that, remove that from the captions, so the lora absorbs that. After, when using the lora, add those tags as they might help force that idea. Other than that just play around with the stuff you can adjust when training. Steps, learning rates, etc. Make sure you have good quality images rather than many images. I hear constantly that good images + captioning is more important than any settings when training.
I was searching for a pitch black skin lora and yours was the only I found.
Wanted to generate some old school drows.
After your lora, since she is the queen of mars, I searched for Dejah Thors, the princess of mars, and was quite disappointed she was not made yet.
PS the realistic generations fail utterly, mostly due to the mouth.
I'm running into a problem when using this, where no matter what checkpoint or sampling model I use, I get a range of ethnically ambiguous women, and can not get it to give me solid black skin like hers. I've tried to copy your example prompts with very limited success. I've tried AnythingV3, Divine Elegance, Vinaigrette, and Aniverse, Any suggestions?
No idea, all generation data is in the images, I used mostly anime models and the only issues I faced getting her likeness is sometimes her facial mark would get cut or she would have a mouth. On realistic models, you'll just mostly get black women, since in the training there were practically no realistic data since she is originally a cartoon.
@wartificialai765 I'll keep trying I guess. Buddy of mine is obsessed with this chick and keeps pestering me for requests lmao
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.