A comfyui workflow for Krea 2 with optional image editing nodes and a switch node to choose between T2I or I2I (it uses the qwen edit conditioning nodes for that)
Custom nodes used in this
INT8-Fast to load the int8-convrot version from here (highly recommended on RTX 30xx)
https://huggingface.co/lilcheaty/Krea2-INT8-ConvRot/blob/main/Krea2-Turbo-int8-ConvRot.safetensors
Replace with load diffusion model if you don't use the INT8 model
ConditioningKrea2Rebalance to ease/remove safety filters, if you want nude men, raise the multiplier to 3
Comfyui-Ease-Use for the styles selector. I like it, your mileage may vary
Description
FAQ
Comments (6)
i have no idea what i am doing here, can you have some more documentation or those note blocks, what am i supposed to do if i am editing, how to keep the face consistent etc or if i am using the "krea2_turbo_fp8_scaled.safetensors" how does it work, it would really help
Well, this isn't an Editing model, so I guess the wf is doing the best it can using the qwen nodes, but for much of the Editing work, I think we'll have to wait until Krea releases their own editing suite in the future.
@ArtiOtaku Exactly, it's not meant as an editing model, but it has some editing capabilities. But if you want exact face transfer, I don't think it can do that
What is the "boogu image turbo lora" supposed to do ?
I use it with and without it and i'm getting literally the same image, i'm missing something ?
Nothing, it's a left over from when I copied that node from another workflow
this model doesn't edit original provided image, it is just redraws it from latent description. Not suitable for edition, incomaprable to flux2 oe qwen because it is just not meant for that.