"Buy Pink Buzz Now! uwu~" - CivChan
Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Behold the power of Anima! These images are pure text2image. No edits. No upscale or Adetailer.
This lora has a lot of custom things trained into it:
p1nkbz- Pink Buzzk1llc1v, text says: "I killed CivBot"c1vtrs, text says: "CivBot Tears"n0tst0nks, text says: "NOT STONKS"text says, "Buy Pink Buzz to Unlock", convenient censoring, censored
And maybe some other secrets!
Description
CivChan, purple eyes, pink hair, short hair,
black dress, maid, white apron, maid headdress, , , yandere, , , p1nkbz, , , k1llc1v, , , text says: "Buy Pink Buzz to Unlock", , , c1vtrs, , , , ,
<lora:CivChan_ANIMA:1.0>,
FAQ
Comments (20)
SUGOI!!! 🤩
Arigathanks lol
Thanks for the inspiration! I've been wanting to make a CivChan lora and also a sarcastic "Fake buzz type" lora for a long time so when you mentioned it yesterday it all came together.
Edit: ROFL thanks for all the tips!!
@CitronLegacy excellent!
I really need to test Anima and this LoRA may have been the needed push :D
I highly recommend Anima! The more I use it the more I am impressed by it.
Also glad you like the Lora! I had a bunch of fun custom things I added to the dataset that Illustrious wouldn't have been able to learn/replicate.
@n_Arno Thanks for the huge tip!! You are awesome!
@CitronLegacy I have been experimenting with ZIT but it is pushing my hardware to much and since realistic stuff is not my main jam, i'll definitely give it a try... the only issue being finding or coding the needed software to run it, since i am still allergic to Comfy ^^;
@CitronLegacy You're welcome! But you're the awesome one :D
@n_Arno Yeah I felt the same about ZIT. I made a ton of loras on it but I dont usually make realistic things. ANIMA is waaaaaay better than ZIT for anime things. I did a few tests and I think ANIMA might do realistic well too https://civitai.com/images/121816167 So its better than ZIT at everything! (maybe 😝)
As for not using comfy, my friend @RisingV got ANIMA working on Forge.
@n_Arno Lol thanks! Well then we are both Awesome! 🤜🤛
@CitronLegacy thanks for the shoutout. @n_Arno I can confirm Anima is working on Forge Neo by Haoming02 since commit b371731. It is a basically a continuation of sd webui forge by lllyasviel, adding support for newer models like Zimage, Flux.2-Klein or Anima. If you want to use it for Anima be sure to install the "neo" branch. Also available in Stability Matrix, if you use that.
@n_Arno Oh yeah, and if you want to try it, do not use schedule type "Automatic" since it might give you a black image. Use "Beta" or another scheduler, I made two post comparing different Samplers and Schedulers with CAT Anima:
https://civitai.com/posts/26655050
https://civitai.com/posts/26656302
@RisingV Forge Neo ask for a more recent version of Pytorch that deprecated my GPU 😭
I'll see if i can find a workaround.
@RisingV Thanks! For the tips!
@n_Arno too bad. I did not know newest pytorch does not work with older gpus. But you can change pytorch version manually it seems.
@RisingV yes, but so many packages depends on this one. I did try changeing some versions but didn't make it works previously, i'll try again! (And i should have get ahold of a RTX 30xx before prices got higher again, that one is on me, i'm stuck with a GTX 1060 for now, with its compute capability 6.1 T_T)
@n_Arno So I don't know much about this myself, but I have different versions of forge installed, which seem to run their own pytorch version (including original forge with pytorch version: 2.3.1+cu121). I think all of them are running in a special environment called venv. You can set up such an environment yourself I guess. But all of the forge versions I am using are installed with Stability Matrix, which already takes care of that.
@n_Arno Impressive you can do sdxl LoRa training with a GTX 1060 (I hope you have the 6GB version :D). RTX30xx seems to be more or less perfect for sdxl and anima also runs good (I have a RTX 3080 with 10 GB though).
@RisingV It took a whole lot of hacking (even if i do know python, virtualenv and stuff) to get it to work (with xformers attention only, sage and flash attention aren't possible with my GPU), but now i have forge neo up and running on my potato setup! I still need to debug some of my favorite extensions that aren't working anymore, but i should be able to test Anima 😊
@n_Arno Oh, that's great! I am not using any special attention function in Forge, but I guess it saves a lot of vram? I am using xformers for LoRa training only. Actually I did not really know what it does until I switched from kohya_ss to sd-scripts this past week and found that my vram isn't sufficient without using it :D
I also noticed I had to change some settings in the extensions with the recent version and also hitting "apply settings" changed about 420(?) settings and fixed an issue I had with an extension.
Well, happy generating then ;)
Oh and my first anima lora will be published later today :)