is dmd2 same as https://civitai.com/models/2466415?modelVersionId=2773137 ?

@reakaakasky I've used the Cosmos-Predict2.5-2B base distilled extracted DMD2 LoRA with v0.23 dmd2 and it's like a match made in heaven like it works really well. It stabilises anima to the best I've seen. It's a bit of a balancing act with getting the strength right to what I like but it's really good. The guy who made it does the hanzogak/comradeship fine-tunes.

Also, there is a great custom node for ComfyUI if you ever want to merge LORAs or use multiple LORAs like charecter and style ones together.

I have tried silveroxides/Anima-Quantized LORAs too. That guy is a great by the way as I think he's part of the legendary Chroma team with LodestoneRock.

https://github.com/ethanfel/ComfyUI-LoRA-Optimizer

However, just for clarification I will never merge your models both LORAs and Checkpoints As I know you explicitly say not to do that and I respect that.

I think Crody one of the top base model creators on there is working on a anima checkpoint too.

I hope also the Incredible KBlueLeaf does something with Anima too. As he's the mastermind who created Kohaku-XL that the now much older but respected Illustrious was built on.

Exciting times for this legendary anime model. Hats off to tdrussell and the rest of the dev team. They are listening to the community feedback to on the official higginface page as well.

Ultimately thank you reakaakasky as all your hard work has really helped make this model be better and more stable I've also learnt about regarding distillation models as well. As it's quite interesting. As I've been using z Imege turbo for a while.

indeed mixing them seems to work well. 0.5 Cosmos + 0.6 RDBT + 1.0 mikkoani + euler is what seems to work best for me.

dmd2 b looks nice

Both are great so far. From my initial tests, "v0.23 dmd2 b" seems to have better color and generally looks better to me. Artist styles / LoRAs miiiight be deviating a bit more but I feel like producing a "nicer" image faster takes priority for a turbo LoRA anyway.

@mac2492 I agree with b having having nicer colors and generally, to me, more pleasing output. This is from very casual and very limited usage.

Prefer b. Cleaner and more stable than "v0.23 dmd2" in my tests.

To me this is the most powerful "creative -> stable slider". I can wait 30s for a good image, the main problem of vanilla anima is stability. Adding this LoRA with 0.2 strength perfectly solved my problem. Or full strength when I want to test something in 5s. Stability takes the priority.