Can the (in)famous anything series generate literally anything? No. They can't generate anything apart from anime styles.
So here you go. An ambitious model that can generate literally anything; from anime style to photorealistic "atashi".
The model was trained based on ChaosMix, with images generated from MoeSharpV1 and "atashi" the artificial girlfriend.
It is NAI leak free, and yet powerful because the basic model is the latest Stable Diffusion 2.1 768.
I didn't upload a yaml because it's no longer necessary when using webUI. If you need one, just download and rename it from somewhere.
Sample images will be added later.
Description
FAQ
Comments (14)
Face too cartoonish in realistic body. Some examples with a realistic face in realistic body ?
Can you upload a smaller 2gb pruned fp16 version (safetensor format) ?
Since it's 2 AM JST, I'll work on the realistic part tomorrow.
I don't prefer creating a fp16 version because I hear it will be less accurate.
However, I will think about creating a safetensor version.
What about the last image I added? Is it realistic enough for you?
https://civitai.com/gallery/122911
@ThePioneer OK. The problem is there are so many great models, but 7gb here and 5gb there, and the terabytes are flying away very fast. So the 2gb versions will be very useful.
@ThePioneer Much better ! The eyes are now more realistic and the face less plastic. Did you just prompted the model or did you use some inpainting/postprocessing ?
@ritcher1 It was generated just by prompting (thus it should work better if you use inpainting, postprocessing, or img2img). As for the size, I'm not sure if I can compress it via webUI, and will check it out (but first off, I need to have a sleep lol).
@ThePioneer Ok, do what you can, of course. I'm interested in this model, but I can't always download 7gb or 5gb models. Many authors have already added the smaller version on request, so I hope to find this one too.
@ritcher1 Sorry for the delay. fp16.safetensors is now available. Enjoy :)
@ThePioneer Thank you very much !
Hello! Thank you for the upload!! However I got an error trying to load the model :
"RuntimeError: The model for Stable Diffusion has not been loaded yet! If you've tried to load it, please check the logs above this message for errors (while loading the model)."
I have a nai config file and I have renamed it to "trulyanything_v1.yaml" and it still fails to load. May be I need the original config file? I'm not using A1111. Thank you very much!!!
If you're not using Automatic 1111, you'll need a yaml based on Stable Diffusion 2.1 768. You might find one on huggingface. NAI is based on SD 1.4, and that's why the yaml didn't work.
Hi!!!
Thanks for the reply. I downloaded the 2.1 yaml file and renamed it. They both loaded successfully. Took about 9 minutes to load. Really! (My models all take about 5-10 minutes to load anyway. Any solution?) and 40sec to generate a totally black image. Nothing to see. LOL. Oh well, it's ok...
@Zenomorph Maybe you might want to check out the nsfw filter, since waifu is notorious for keeping it, and it might be inherited (I'm not really sure about it though because I usually run models on A1111 via colab, not locally).
You could try finding a way to remove it on your environment.
If you once convert the model to diffusers, the following line might work;
The solution to load your models faster would be to either use Automatic 1111, have a better GPU, or use a service and run it online (just like I use colab).
It says truly anything, but all i see are only semi realistic girls, specifically asian girls.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.