Works great with CacheDit(x2 speed)
Exactly 2x speedup after default 3 wakeup step for example.
Just the subj, for info to others—maybe some don't know you can speed up generation with CacheDit.
Overall, an excellent model, thanks!
What custom node and which settings? I get very noisy images.
What custom node and which settings? I get very noisy images.
This custom node.
I found a link to the repo here:
https://www.reddit.com/r/StableDiffusion/comments/1p8v52y/zimage_cachedit_in_comfyui/
Unfortunately, this specific implementation has been removed for some reason, even though it supports all the models I know of that I've tested—over a dozen base ones, including some good but not very popular ones like Kandinsky 5.0. I haven't come across a model yet that this implementation doesn't work with.
All settings are at default, just like in the screenshot.
The images are absolutely clean, without any noise
Wow.
X4 faster + with this LoRa(CFG 1). ><
That's a super great Lora and checkpoint by that guy. It really adds more details, stability and prompt adherence to anima even on the default 4CFG and 20 to 30 steps. Highly recommend. He has a Illustrious, z Image variant too.
The dropbox file appears to be the Jasonzzt CacheDiT which gives me very noise output for anima and is not what is shown int eh screenshot
The dropbox file appears to be the Jasonzzt CacheDiT which gives me very noise output for anima and is not what is shown int eh screenshot
Sorry,my mistake.
A cacheDIT github from screenshot is https://github.com/rakib91221/comfyui-cache-dit
even with this repo I get very noisy results, not sure how OP got good results, tried default variables + a lot of combos
Wow.
X4 faster + with this LoRa(CFG 1). ><
Too bad that project for Anima is abandoned because of the unfriendly license. But very interesting experimentally.
A cacheDIT github from screenshot is https://github.com/rakib91221/comfyui-cache-dit
even with this repo I get very noisy results, not sure how OP got good results, tried default variables + a lot of combos
What samplers+schedulers? Some variants give noisy image by default
As u see no any noisy(with workflow),i think this is your side some strange issue.
https://disk.yandex.ru/i/Xu88V3llZHnvtQ
And a lot more samplers testing with Simple scheduler(a used Add delay node in workflow because generated on Ryzen APU and have core temperature issues,in any case it doesn't affect anything)
https://www.dropbox.com/scl/fi/acf9yu2frphm4v6tdxeec/Anima_Samplers.7z?rlkey=pqkrdfonfsljl7yktey556sp2&st=nmwgojjf&dl=0
Maybe you're using some favorite samplers and schedulers that doesn't work at all with Anima, and as you can see, there are quite a few of them.
