Krea2TurboFP8
Description
FAQ
Comments (53)
Loving it so far <3 Does a eclectic, wide range of strong, distinctive styles right out of the box.
Captioning is also flawless from what I've tried.
Character likeness is also very good with the base model (on nothing more than just a single character name; no description even).
good at anime, i wish some NSFW lora for krea2
We need a fork of this one with fully NSFW capability XD
@smoonHacker It works very well with nsfw tho just try the prompts
https://civitai.red/models/2725430/krea-2-nsfw-v3?modelVersionId=3071760 (Doesn't change the image you started with too much).
it could be the next flagship model, still need a easy way to uncensor
It is uncensored. Watch this video, it shows how to get rid of restrictions. And the purest truth in that video - I tried it myself yesterday. I didn't try anything 'hardcore' because I'm not interested in it. But it produces the usual aesthetic 'nsfw' images perfectly.
It's an interesting models that handles both realistic and anime pretty well with the correct prompt and direction. Details on clothes and structures have a good definition, patters are defined properly.
Guys the model is already uncensored make your prompt just detailed
some example please, it wont even gen woman with chest bigger than B cup
nsfw content seems very deep and difficult to trigger successfully
its a hit or miss with the uncensored stuff, just give it more prompt details and it should do it.
Some words or phrases are definitively filtered.
@xhamster Add "conditioning krea2 rebalance" node and you should get chest bigger than head (if you want it). Simply connect this node from text encoder positive to ksampler positive. And with spicy images would not to be any problems. Not so deep, looks like in tuned models, but Krea2 work "out of box".
@herkus_baronas631 nah that node is useless stop using it, it destroy the model
@xhamster This node working and nothing wrong. I today used Krea2 wf with this node. About example. Where i can load it? To civitai?
@herkus_baronas631 but toooooo many nsfw lora, so dont use the rebalance node
@ping26 I no need 'hard style' loras. i need just some slender curvy womans, if making images with them. So for me is enough this node. Or one unlocked model for 'playboy' style images.
This custom node removes the built-in Krea 2 safety filter. https://github.com/nova452/ComfyUI-ConditioningKrea2Rebalance
the idea is right, but the params are strange
it does work but the output seems drop a lot of quality and strange coloring
@xhamster Yes, I noticed that too, and I noticed that the texture of the skin feels more plastic.
i agree it works but the quality drops and character knowledge also gets lesser.
for me works perfect
How do you use it?
@chaank2013585 Connect this node to the positive prompt node and output node in ksampler.
Does this model have enough potential to overtake z-image-turbo in the long run?
Yes, maybe yes. The details are far away from Z-Image Capabilities.
I think yes. "From the box" Krea2 for me give better results than Z image of Klein. I made maybe 15-20 images with Krea2 turbo and all results was good. For NSFW... Krea accidentally created such an image. And as far as I could see, Krea has been trained on the anatomy of the upper part of the female body. I wondered - is it possible that the text encoder simply does not pass the request to the model. I am not an expert, but maybe after applying the corrected text encoder, the model itself will not even need to be trained. I don't know, these are just observations. But I really liked the model. I would say its photorealism is much higher quality than Zimage or Klein.
PS. This model handles text well. I tried English, Lithuanian and Russian texts. The model correctly represented both Lithuanian "žšąų" and Cyrillic. It even coped with bilingual text - Russian text on one line, Lithuanian on the other.
I'd say yes.. this is an 0.1 model, and it already rocks, we can just imagine what future versions will look like.
I'm a big Z fan but damn this is a good model and out the box its doing better things than what Z did at least for me. can't wait for things to come for it.
@MrGhost_27 @MrGhost_27 I spent a couple of hours yesterday with Krea2. The results were pretty much what I wanted. No 3 arms, no 6 fingers. It did a lot of things better than other models with special Lora's.
And if this is only version 0.1, what can we expect from version 1.0?
What about an IP-Adapter? z-image-turbo doesn't have an IP-Adapter or edit model right? It would be nice if Krea 2 did.
@herkus_baronas631 yeah the text is quite incredible, at first I thought that I was just getting incredibly lucky
Any 8GB vram users here? Does this work?
Yes
yup, works well
Flawless even on a 6 GB RTX4050
It seems that this model has the same issue that Z-Image had when it first launched. All the generated images look very similar. Is anyone else experiencing this problem? How can it be fixed? (using ComfiUI template)
Hi! Unfortunately, the base model doesn’t shine in terms of variety—but that’s expected, since it’s meant to be a foundation. We’ll have to wait for users to release their own fine-tuned versions, though that won’t happen right away. Usually, the first or hastily made community versions end up worse than the base model—everyone’s rushing to be first, and many don’t care much about quality.
In my opinion, it’s best to wait. This model has already proven itself superior to Z-Image in many aspects.
it is because of the turbo distillation, try the raw model and you see there is much more variety and dynamic lively poses. Turbo comes at a cost. https://huggingface.co/Comfy-Org/Krea-2/tree/main
and yeah i find Krea2 much better than ZImage, QwenImage 2512, Flux.2 Klein 9B.
I found this custom node. It work pretty well on a "Flux Preset" in Bold mode https://github.com/RamonGuthrie/ComfyUI-RBG-SmartSeedVariance
ER-SDE / Simple and noise 0.7-0.8 seems like a sweet spot for realism.
Training LoRa on BF16 Raw without transformer quantization is pretty straightforward, and produces excellent results in Turbo FP8 generations.
does this work with Forge Neo?
Not (yet), but support is planned.
It's supported now
Where are the vae's, text encoder, and basic worflow?
Update ComfyUI. Open template, search "krea2", load workflow. If you get errors, copy the model names, Google them, download
Is there any comparison with FP8 turbo by Comfy? Why do we need this custom version?
so i must download this one, or the Huggin face one
Ok so what is this, there is already a Krea2 FP8 scaled from Hugging Face (ComfyUI) which is 13GB in size, this is 12GB. Author can't bother to write even a sentence of description. Sorry but this should be taken down for not even bothering to tell anyone what it is -- at best.
The model is quite good! It understands Prompt perfectly and generates quickly. You can skip LoRa if Prompt is very detailed.Good job!