ATTENTION THIS IS IMPORTANT!!!! DUE TO HOW THE FILES WERE STRUCTURED MAKE SURE YOU PUT "FULL BODY" IN THE PROMPT FOR THE BEST RESULTS!
THIS MODEL WAS TRAINED ON MEINAMIX V9!
Example prompt: vtubermodel of 1girl , dark brown hair ,orange eyes , fulffy weasel ears , high quality , extra detailed , rendered, (green hoodie:1.3), white background, full body, smiling, semi realism <lora:VtuberMaker:0.7>
Neg prompt : (worst quality, low quality:1.4), border, frame, (large breasts:1.4), watermark, signature bad eyes , bad hands, asymmetrical, complex background , black background, extra fingers
this model was trained with the meinamix v9 model
Description
THIS VERSION WAS TRAINED ON SD 1.5 AND IS INCREDIBLY POTENT
Example prompts for this version :
vtubermodel of 1girl , brown hair , light brown eyes , (green hoodie:1.3) , fluffy weasel ears , white thigh highs, high quality , extra detailed , rendered, (white background:1.2), full body , smiling<lora:VtuberMakerV2:0.5>
Neg prompt: bad eyes , (bad hands:1.3) , extra limbs , extra fingers , assymetrical , low quality, complex background , background elements, bad anatomy
FAQ
Comments (13)
Love the preview pics, would love to see more example pics too. I noticed it's tagged as LoRA, should this be tagged as a checkpoint?
hey i updated the model im new at this and yeah it was a checkpoint accidentally oops .... well now its a lora!
@HorridWeasel are you sure, the file size is way more in line with a LoRA?
@Cjb what youre seeing now is the revamp! i had ckpt files in there by accident but everything is now fixed i personally recommend v2.0 as it has more stylization flexibility ! enjoy :)
@HorridWeasel friendly reminder that generally you'd only want to train on the nai model/ anime-full-pruned. You want a good database that is pruned towards anime, got a lot of concepts, scenery data and styles, while not overfitting your lora. Training on sd 1.5 is just..
notice how your first pic has lora strength 0.3 meaning it's barely activated, while your second is already overcooked at 0.7. You can aim towards 0.8-1 strength, cfg 7, Karras++ sampler
@cyan8 heya thanks for the suggestion! ill make sure to train v3 the way you suggested ^^
@cyan8 i tried your suggestions and they didnt work i just personally believe different loras require different offsets to work , have a nice day ^^
@cyan8 if u have anymore suggestions im open to hearing them im using kohaya gui!
@HorridWeasel Thanks for sharing, can't wait to try your LoRA out!
@ThePotato one tip if youre noticing an overbake try not using the keyword!
Hey! Was it trained only on female avatars? Is it possible to generate male avatar with that lora?
heya yes male models were trained too
Looks interesting, but can it somehow be used to create a working model?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.