There are two files. The bigger one is the AIO, the "pruned" file is the diffusion model (you will need TE and VAE). You can find them on link bellow.
Next Versions
Adding improvements and my style.
-Theta
V1 and my style.
-Alpha
This is my own version, after a training day. Now we don't need use improvements tags.
Still is working as preview, tags or natural lenguage.
Enjoy it and more improvements will come.
-Preview.
Go to the official Anima Huggingface repo
Remember use masterpiece, best quality, score_9 unless
I only merge all to use it like a AIO checkpoint.
Description
Preview 3 based
FAQ
Comments (13)
Amazing AIO Anima model. Thank you for creating it! P.S. Works on Forge Neo :)
does this lock in the perfectRSB style in every generation?
having to use artist names are becoming tedious and they never yield what is expected. especially when you dont care for knowledge of who's style is who's
If it is the style it is constant the same. At most you can change it to ask for a specific style such as dragon ball, kimetsu no yaiba,.... I had the same problem and I always want to use my own style like 2.5D and modify it with more or less realistic tag.
@frakerkill_ I am gonna use it today. i trained some loras to test (still figuring out the covergence vs lr vs step count with a problematic dataset. but its the same dataset I used to make loras thats worked best with all prsb pony and illustrious versions which just got better every new version.
those used to give me the best results that had your signature style. one thing i noticed is that your pony and illustrious model forced your style onto the image, while allowing my lora to handle shape and textures transfered but kept your style's coloring. cant get that anywhere else, or very difficult to.
@ttaetherai I always try to made model that can adapt to other Loras and "how to say it" add into them to take improvements from them. It's like distilled good alcoholic drinks 🤣
@frakerkill_ tested this anima version.
to explain my problematic lora dataset
my character was a 3DCG OC i designed years prior to AI. purely 3dcg dataset.
the anatomy, is... well, mine, so Photos.
base anima always forced a burnt 3DCG with mangled realistic textures. cos sure, its not a Z-image thing
as expected - yours DOESN'T. works exactly like pony and Illustrious. your style and colors are first, my shape 2nd and my textures last. as it should be.
Thank you for your service
@frakerkill_ tested this anima version.
to explain my problematic lora dataset
my character was a 3DCG OC i designed years prior to AI. purely 3dcg dataset.
the anatomy, is... well, mine, so Photos.
base anima always forced a burnt 3DCG with mangled realistic textures. cos sure, its not a Z-image thing
as expected - yours DOESN'T. works exactly like pony and Illustrious. your style and colors are first, my shape 2nd and my textures last. as it should be.
Thank you for your service
@ttaetherai I want see, some examples. If you want publish it
@frakerkill_ Yeah Sure! I just need to pull out my working prompts from my pony and illustrious days and make some.
@frakerkill_ https://civitai.red/posts/28150342
for the life of me, I cant understand why the workflow didnt transfer with the upload. which forced me to leave out the seed and the sampler reads euler instead of er_sde. i'll send one randomly in dm so you can see the workflow if it interests you, its a work in progress stripped down z-image turbo workflow that emulates the old a1111 pipeline. generate->hires fix with upscale model->ultimate upscale using seedvr. i couldnt get a decent adetailer model to work so i left it out
@ttaetherai I use imagesaver comfyui for keep the metadata inside the images
@frakerkill_ no, i do too, i think its something wrong civit side
Guys:
Download this one
Your loras will work ✔️
You wont get style bleed through if you dont use @artist ✔️
The images are clear and consistent. ✔️