Create Colorful Anime Treasure with CAT!
Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
What is Anima? Anima is the first ComfyUI sponsored model! Please read the official announcement from ComfyUI here.
Please check the official repo important details about how to use the new base model: https://huggingface.co/circlestone-labs/Anima
You need
Text Encoder: qwen_3_06b_base.safetensors
VAE: qwen_image_vae.safetensors (Actually I baked in this VAE but I haven't tested without it yet.)
AnimaCAT_v1 - Anima-Preview x (RoxyRibbonStyle_ANIMA_Lora:0.3) x (Unreleased_MoonlightCAT_ANIMA_Lora:0.5)
Legal
The CircleStone Model is licensed by CircleStone Labs LLC under the CircleStone Non-Commercial License. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Description
FAQ
Comments (17)
I made a test with different schedule types available in Forge Neo for ER SDE sampler and scheduled a post. I'd rate results by prompt adherence/quality in the following:
best prompt adherence:
Beta
Bong Tangent
good prompt adherence:
Normal
Simple
SGM Uniform
DDIM
FlowMatchEulerDiscrete
ok prompt adherence/some inconsistencies:
Linear Quadratic
KL Optimal
blurry/bad anatomie:
Karras
Exponential
Polyexponential
Align Your Steps
black image:
Uniform
Turbo
For visual quality, see for yourself, but I'd say it is good for the first two categories. Don't use schedule type "Automatic", it gives you a black image (at least for me). Test was done with 30 steps, took 21-23 s for every schedule type, so no big difference. Maybe the schedule types in (blurry/bad anatomie) category need more steps, have not tested that.
Thanks for your diligent testing! Historically I use Karras so it feels a bit weird to switch to something else. I've only used simple but I will give Beta a try!
On a slightly different topic: I'm really impressed by how clean all the images look! Anima seems to have no problem making very detailed shoes and also keeping the rest of the image looking good. Maybe I am biased but Anima just seems a bit more precise than Illustrious.
Also the background on this turned out surprisingly clean https://civitai.com/images/121340625
@CitronLegacy Well, I try to find the best settings for myself, so why not sharing them? ;)
I have been using Euler for some time now with Illustrious and scheduler set to "Automatic" so I don't really know which scheduler it was using.^^
However I think Beta is also the standard for Zimage?
I think I used DMP++ 2M Karras, when starting to make images with pony :D
Yeah the test images look really good (well with some execptions). But I haven't made much more so I can't really tell. I also made another bunch of these images with different samplers this time, but the post seems to have gotten lost^^
Here its is: https://civitai.com/posts/26656302
Maybe the background is so clean bc they also trained on non-anime stuff (LAION-POP)?
Flux 2 sampler gives good result, may be best🤔
@mifink94 Which sampler is the flux 2 sampler? Euler? Or are you reffering to some custom sampler for flux 2?
@RisingV oh, sorry, scheduler... i means scheduler. (Euler a sampler+flux 2 scheduler). SGM Uniform is good too. black image because "automatic" use Karras. Karras not working.
@mifink94 ah, ok. Mmmh, did not try flux2. Maybe I forgot that one?
In my test with ER SDE only Uniform and Turbo schedulers resulted in black images.
@RisingV hmmm, magic, whot interface do you use?
I have tested it and I can say that the baked in vae work..
Awesome! Thank you very much for testing it! I always prefer Checkpoints with baked VAEs so I'm really glad its working.
Oh yeah, also made a vae vs. non-vae test: https://civitai.com/posts/26639478
It's working without, but looks a little different.
@RisingV Thanks for testing! The version with VAE is better, because the see-through sleeves are a nice touch, but yeah they are 90% the same. Thanks for sharing the comparison so we can use it to choose what we prefer.
N/uhh.. I should be the one thanking you for the amazing model , really great work I got really high quality stuff out of it so thanks (really much!!)
Im quite conservative, but this thing pleasantly surprised me.
thenks a lot!
Boss, do you try upscaling images by anima?
Thank you glad you like it!
I haven't tried upscaling but Image to Image is really nice and cleans up images pretty well. I'd bet that upscaling is good.
I wonder when they will release the full model?O_O
What is "Zel image saver"? Why are there so many missing nodes when I try to load generated images?