I'm very excited to bring my first LoRA. In this case, I created an anime character by drawing it myself, then used A1111 to generate more images of this girl with different poses and expressions (blushing, angry, tsundere, orgasm). The model is trained with both SFW and NSFW content. Finally, my recommendation is to generate the images in Pony (768x1024).
Trigger words: nora_habana, black hair, red eyes, freckles, open jacket, bra, thong.
I recommend using weights between 0.6-0.8 (if the description is complex or other LoRAs are used to complement) and using fewer than 75 tokens.
Description
I'm very excited to bring my first LoRA. In this case, I created an anime character by drawing it myself, then used A1111 to generate more images of this girl with different poses and expressions (blushing, angry, tsundere, orgasm). The model is trained with both SFW and NSFW content. Finally, my recommendation is to generate the images in Pony (768x1024).
Trigger words: nora_habana, black hair, red eyes, freckles, open jacket, bra, thong.
I recommend using weights between 0.6-0.8 (if the description is complex or other LoRAs are used to complement) and using fewer than 75 tokens.
FAQ
Comments (7)
I would greatly appreciate it if someone could explain to me how to make the LoRA take up less space. Normally, it takes up a maximum of 70MB, but this particular one is very large. Does anyone know how to fix it without losing precision in the LoRA?
I hope you find a useful tutorial or get a helping hand. Looking forward to seeing this cut down to a tighter size.
@TheFatedCucumber I've been reviewing documentation from other creators, and to reduce the file size, it is ideal to lower the "Network rank_dimension" during model training (though this decreases precision). I will re-train the LoRA to see how the results turn out. Thank you very much for your interest and comment!
@Charlytos Thanks for your own time on these kinds of projects and sharing them. Looking forward to seeing your next pass and the file size change results.
@Charlytos I remember seeing someone discussing this and saying something on the lines of "train it larger file size first then reduce it with kohya_ss and that won't reduce accuracy" (I'm sure to some degree, but not as much) but I'm not sure if I remember it correctly. I haven't even manages to train a LoRa worth using yet, so I'll leave that to the others, not sure if you've seen anything similar to that as well or not though.
@Kaladae I´ll try it. Thank you so much.
@Charlytos Let me know how it turns out. Good luck!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.