This simple workflow for Krea 2 Turbo allows you to use 2 distinct character loras and generate an image/scene with them both in it. Usually if you try this in a standard workflow both character end up looking like weird hybrids of each other. This worklow can still experience some of that, but it's a good starting point to get around that. Just follow the notes on the workflow where to load your male/female character loras and where to input their corresponding trigger words.
Tested With (for nsfw):
Lustify V10 (Krea 2) https://civarchive.com/models/573152/lustify-nsfw-checkpoint
and krea2_turbo_int8_convrot.safetensors + KNPV4.1_pre (1).safetensors
https://civarchive.com/models/2725430/krea-2-nsfw-v4?modelVersionId=3085473
(For sfw) It should work too if you try the standard krea 2 model with no loras or just sfw ones.
There are probably other and maybe better ways to do this, but I've been surprised at the results I can get with this workflow. It's not perfect. There's some identity bleed, but if you follow the notes that can be reduced.
This is still more fun than generating an image with a single character lora, then trying to come back and mask and regenerate with a second character lora. I've yet to find a great solution for doing that which isn't super manual and full of trial and error.
This works on my 8 gigs vram 32 gigs ram setup.
How I trained the Loras easily Online since I am GPU poor (costs a few dollars):
To train my loras I used pixaroma's tutorial: https://www.youtube.com/watch?v=XyMuTsDsvEY . Follow the steps for a character lora on fal.ai, and use pixaroma's nodepack and krea lora conversion node to convert the output to work with comfyui and krea 2 turbo.
Description
FAQ
Comments (6)
This workflow is pretty straightforward but lacks critical features for producing high-fidelity outputs. For instance, it’s missing ADetailer-style nodes for localized inpainting and refinement of specific regions. It also relies on the default ComfyUI LoRA loader, which is suboptimal when you want to use multiple LoRAs or when granular control over their influence is needed. A more advanced multi-LoRA loader like Power LoRA loader would offer far better flexibility and precision. The same limitation applies to the default VAE loader. This becomes a bottleneck if you want to experiment with alternative VAEs compatible with Krea 2, such as the WAN 2.1 VAE. For beginners, this setup is acceptable, but it won’t deliver the same level of quality or control as more sophisticated, production-grade workflows.
nice "review" written by AI. didn't see any workflows made by you, mr blowhard.
The other guy with his AI review was clearly not helpful so let me give it a try.
Sadly i can't say that this is working. Regional prompting means something very specific in my mind and what is done here is not that.
This is in essence the most barebones workflow you can have and having 2 lora nodes does nothing aside from having the illusion of separation.
Try this workflow with two female loras and see how this does not work at all. Having many demo images where the face of one char is hidden feels like a sneaky trick.
I wish it would work but it does not, sry.
Yea no offense taken. I'm new at building workflows and this is meant for beginners.It's a simple workflow that works decent for me so thought id share it. made it with one 1 male/ 1 female in mind.
Specifically worked the best when I made loras using pixaromas (more info on the workflow) tutorial and settings.
It requires a lot of trial and error and prompting. The prompt structure when followed can help a lot to keep the 2 characters separate. Make sure the lora trigger word are plugged in multiple times and their placement in the scene is mentioned. works best with the lustify krea 2 fine tune which is still in early access and should be out for free soon. Adjusting each Lora strength and the strength of the NSFW lore makes a big difference as well.
I may add onto it more in the future
Also just wanted to say I in no way tried to be sneaky with the helmet character. I tried to pick loras on civitai so people could compare the characters to see that it works and so they could test it out with all the loras. There are very few character loras on civit ai for krea2 right now. And from what I've found it's normally very hard to keep male and female parts separate when using a male and female lora. So really the main benefit of this workflow is the prompt structure so people can see how to fix that a bit. It's the nature of krea2 and it doesn't like having 2 character loras at once. theres some experimental krea 2 edit loras out that are promising but they also have a lot of quirks.
You are right on the term regional prompting. for sdxl, which i used the most before this I thought that was the term for the text prompting for regions, but I see it means someting different . Changed the title to spatial prompting which google/ai says is the preferred term for this.