This is my first time training a model, but i hope you enjoy:
if you're using the v1 dont forget to use the tags: full_body, virtual_youtuber, standing.
It works with anything 3 to 4.5 and models that use a similar dataset.
if you're using the v2 version i recommend you to use the MeinaMix and you don't need any tags, but if you model is being cut use "full_body", and depending on the tags you use for the characteristics you may need to add "straight-on".
if you need help feel free to send a msg on the discord:
https://discord.gg/mZpWrdcZVz
Description
Just a update for the model, this version is making more coherent models.
This time i trained the hypernet using the MeinaMix model, so i recommend you to use the same:
https://civitai.com/models/7240/meinamix
Now you are not obliged to add the tags virtual_youtuber, standing, fullbody. But if you model is being cut use "full_body", and depending on the tags you use for the characteristics you may need to add "straight-on".
BR --
Esse é um update da versão anterior, ela consegue criar modelos mais coerentes e erra um pouco menos nos detalhes.
Eu treinei a hypernet usando o MeinaMix, então eu recomendo vocês usarem o mesmo modelo, não tenho certeza se funciona em outros.
https://civitai.com/models/7240/meinamix
Não é mais obrigatório usar as tags: virtual_youtuber, standing, fullbody. Mas se você ver que seu modelo está sendo cortado, pode querer usar full_body, e dependendo das tags que você coloca para as características do modelo, pode querer usar "straight-on" para ajudar ele a ficar na posição.
FAQ
Comments (2)
What resolution/aspect ratio would be recommended for this? i keep on getting bodies that are too squished or legs with 4 segments
Hexzees you can use 256x256, 256x512, 384x786, 786x786. If your model is a male or has long ears like a bunny, try to use 384x1024. When you Hires. Fix, if your denoising is above .8, you may wanna increase your CFG.