For anyone who has not noticed yet, new model versions of Anima have been released by Circlestone Labs. The turbo version seems to give decent results, at least in the first test images I did. Also onsite generation with anima-turbo is only 6 buzz at 8 steps with default settings now (I paid 9 buzz testing yesterday, but the license fee has been lowered to 2 buzz now). So if you just want to try some lora or something, this is a viable option. Some artist styles may not work as expected though.
Also two other versions have been released, aesthetic-v1.0 and aesthetic-v1.0b. I just did a few images with aesthetic-v1.0 and I need to do further tests, but I am not very convinced by it right now. So I guess I stick with anima base for making showcase images.
On another note I finally made Krea 2 work with Forge Neo. At first I was just getting black or distorted images, but no error message. Turns out you need to disable sageattention to make it work. Haven't tested it much, but first results look good. With my RTX 3080 the int8convrot version seems to be the best version to use, even though it does not fit in the VRAM it is still faster than nvfp4 version. Since int8 is natively supported on 20XX/30XX cards, int8convrot seems to be perfect for giving these older GPUs an inference speedup.
The cover image was generated using Krea 2 for the base image and anima for a little upscale and giving it a bit of a different style flavor.