Yo, just grabbed the Dark Beast Krea2 attack version GGUF, params are straight copied from the official ones for now.

The Forge / 工坊 Models / LoRA / 模型与 LoRA
1

Sharing the GGUF version of Dark Beast Krea2 Aggressive Edition. The description clearly states this isn’t the creator’s own merge—the original merge was done by @AiMetatron, they just converted it to GGUF format. Pretty respectful of them to give credit where it’s due.

Recommended settings are ER_SDE or Euler sampler, Simple scheduler, CFG=1, 8 to 16 steps. CFG as low as 1 makes sense for this type of distilled model. Haven’t downloaded and tested it yet, so not gonna hype the results until I actually try it. GGUF is VRAM-friendly, so those with smaller cards should keep an eye on this.

2

Damn, CFG 1 is that low? First time I’ve seen that, kinda curious.

3

Good on them for crediting the original creator.

4

GGUF saves VRAM, a godsend for us 8G users.

5

8-16 steps to generate an image should be pretty fast.

6

Is ER_SDE the perfect match for this? Someone confirm.

7

Converting formats counts as getting work done too, don’t hate.

8

Tested it, ER_SDE really is more stable than Euler, less noise in the shadows