LoRA training cost Civitai Buzz. If you like my work, please consider tipping me so I can keep creating them. Reacting to my images, posting your images and using Civitai generator with my models can also help.
Description
FAQ
Comments (16)
Are you going to do Erika and sachi?
no, there are already good LoRAs for them here so i am focusing on other animes
https://civitai.com/models/74250 https://civitai.com/models/90911
@andinmaro146 Also how do you make these characters look so good? On stable diffusion?
@Suluman45 the LoRAs trained from anime screencaps and i am just using models that produce good anime images and negative embedding https://civitai.com/models/71961/fast-negative-embedding . you can copy the parameters from the sample images and you should get the same results
@andinmaro146 I used google collab to use stable diffusion.
The thing is when I use images from civitai. They aren’t perfect but they look like the character.
However when I use images I find on galleries or google, and I train them, while they do look like the characters, the characters aren’t pure color and occasionally not perfect. Also I can’t always create two people.
What is negative embetting? Does it have anything to do with negative prompt?
@Suluman45 i can create two people but i get bad results, only solo/solo focus images are good and only if you don't use character LoRA because then you will have two same people. if you are using google images make sure the resolution is at least 512x512 when you train LoRA. also if your dataset is from anime screencaps i suggest training and generating with this model https://civitai.com/models/84586/aam-anylora-anime-mix-anime-screencap-style-model
negative embedding is something that you download and then use in the negative prompt and it is making sure you won't get very bad results. you put the file in embeddings folder if you use automatic1111 and you can copy the negative prompt from my sample images
@andinmaro146 automatic 111? Also do You paste negative embeddings in the customs url along with the url of the character downloads?
Also I did download this image but it didn’t look very good.
@Suluman45 You download the negative embedding from civitai for example this is the one i am using https://civitai.com/models/71961/fast-negative-embedding it is not included in the LoRA
in automatic1111 i mean the stable-diffusion-webui project https://github.com/AUTOMATIC1111/stable-diffusion-webui/
i don't know how you generate images but if you using automatic1111 and the same parameters you will get the same result. you can also try online service like pixai to see if you get the results you want
@andinmaro146 so I have to include the negative embetting on the prompt?
I am using automatic111.
It’s still not working.
@Suluman45 yes you add it to the negative prompt like in the samle images after installing it. you can follow the instructions here https://stable-diffusion-art.com/embedding/
@andinmaro146 The thing is even when i don’t download it, because your image already has it, the negitive embedding is already there on the negitive prompt
@Suluman45 yes but it can't work without the file. when you put it in the prompt it using the file just like LoRA
@andinmaro146 The problem is that I put the Lora and negitive embelling in the url. But I can’t see to find the negitive embelling in the Lora?
@Suluman45 it is not in the LoRA it is different thing not related to LoRA. which url you talking about?
@andinmaro146 customs url the link is here. And I am trying to create hiro
https://colab.research.google.com/drive/1wEa-tS10h4LlDykd87TF5zzpXIIQoCmq#scrollTo=8Uvqba0yKW7U
@Suluman45 i don't use colab for generating images so i am not sure. from this notebook i see that if you put the civitai link for both it should put the files in the correct folder but i don't really know if it work. maybe you should ask in civitai discord
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.