Based on 120 niche photos. It's not great, maybe I'll redo it sometime, seems to give a reasonable result for 1 in 8 renditions. My workflow is to generate the image using an anime model for flexibility of composition, then switch to a semi-realistic model at 20%. Finally I scale-up with the same anime model for 20% and refine with a realistic one.
Results are very dependent on the models used so try around, don't write it off if you end up with teddybear on teddybear action, or some hybrid girl-bear.
Recommend keyword (tb1:1.1), anything else is for your experimentation.
1girl, (stuffed toy:1.3), (stuffed animal:1.3), (teddy bear:1.1), (nude:0.7), bed, indoors, solo, hug, bed, barefoot, long hair, lying, breasts, nightdress, (sex with teddy bear:1.1), (1boy:1.2), (best iality:1.3), female masturbation, dark-skinned female, (hetero:1.2), lingerie, pussy, looking down, arousal, aroused, orgasm, looking pleasured, dildo, sex toy, oral, fellatio, lingerie, straddling, girl on top, sex, vaginal,
Try putting animal ears, bear ears, fake ears etc. in the negative prompt to avoid a furry female.
If you get good results, I'd appreciate your posts, thanks
Description
FAQ
Comments (6)
Hello, this one seems interesting but first of all I noticed Lycoris size is really small. Well anyways I downloaded and tried it. It gave me an error on console, are you sure this file is working?
My file is the same size as the uploaded one, and there is definitely a difference in the image composition with or without the lycoris, when using the same prompts. I don't know enough technically to explain why it's small, but definitely working for me. I will try to upload again in case it's corrupted.
Also, it is a Loha, not a lycoris. Not sure if this makes a difference to file size
@Grover0 Today I tested this one once again and suddenly it doesn't give the same error as before. It seems to work but at least for now seems very hard to get any decent results, have to test it a bit more. It's a Loha yes, me neither. 😊
Edit:
[LORA] LoRA version mismatch for BaseModel: E:\stable-diffusion-webui-master\models\Lora\NEW\TB1_loha.safetensors
it gives this one on console log, I just didn't notice it first today...
@Mr_BlackSheep It's trained with a semi-realistic SD 1.5 model. Unless the upload was corrupted I'm out of ideas. Might try training as a lora or lycoris sometime
@Grover0 Well the Checkpoint shouldn't be a issue, all your other stuff seems to work fine but I think this one just doesn't? File is ok, its not corrupted, I was comparing its size to other Lohas and they all seem to be quite same size as LoRas, so it is a very tiny one compared to those.