@llikswonskcalb Thanks for your reply! I wanted to try your workflow and I looked up Krea2 to try it, I get that it's resource heavy, meaning my 3060 12gb isn't really enough? Or maybe I'm doing something wrong?

@llikswonskcalb Alright thanks mate o/ that's unfortunate

@EXEbeast830 haven't tried this workflow yet, my 12gb 3060 setup can run krea2, but I've got 32gb ram that helps compensate for offloading parts of larger models.

If you've got the same, then I'd look into what quant of 'clip/text-encoder' model you're using in case it's chewing up your memory.

If you're using comfyui, there are ways to unload resources that aren't currently in use. It is pretty tight to run it though.

Been a few weeks but I've used some int8 Krea2 models, and think I got between 28-40 seconds per gen after I'd gotten everything loaded for the first run.

@civitainsfw196 I found it weird that I couldn't run Krea 2 and dug some more, since I have a 12gb 3060 and 32 ram as well (luckily before the crisis).

I found a neat tutorials on youtube that helped me a lot to adjust a workflow onc comfyui, since I honestly don't know anything about the technicalities. I know it is not perfect, but I can generate one image in 40-70s, unless I use big loras.

I use qwen3vl 4b fp8 scaled. I know there are others, and I was about to try a 8b convrot, to try if it will slow down the generations. I also use a krea 2 T enhancer but don't know if it changes much.

In any case, thank you so much for your precious comment.