Thanks to all of you making GonzaLomo Krea 2 such a popular model, it is now the flagship model used by generateporn.ai! If you are willing and able to support me some more then I would very much appreciate it if you purchased a monthly or yearly subscription - and paying with crypto would be a bonus as well. I don't charge for this model so even subscribing for one month at the lowest level would help a great deal.
Note: There are two FP8 download options for many versions. The 17GB option is an all-in-one checkpoint that includes the text encoder and vae as well, so load it like a regular checkpoint. The 12GB option is just the unet so load it like any diffusion model.
So many new model architectures, I can't keep up. Krea 2 is definitely going to have a place in my toolbox though. It might be the best current T2I model.
For best results, follow the recommended suggestions in the About This Version section.
v1.0 seems to be very good at NSFW content but it has the problem of producing unprompted nudity.
v1.0 Alt has addressed the unprompted nudity issue to some extent but is weaker at NSFW.
v2.0 Training Base is my test bed for new training.
v2.0 RC1 is a potential release candidate that builds on Training Base.
v2.0 Final is based on the original v1.0, abandoning the ALT, Training Base and RC lineages completely.
v3.0 is just a refinement of v2.0.
Beginning with v2.0, there are at least 5 versions available for download:
BF16
FP8
INT8 convrot
Q4_0 GGUF
FP8 All-in-One Checkpoint (text encoder and vae included)
Description
While I really, really like the v1.0 model, it has some serious flaws that make it unsuitable as a general purpose model. There's a lot of unprompted nudity and some other artifacts that get triggered by the prompt.
So I went back to the drawing board and remixed all the ingredients to come up with an alternative version that can still be used for most content but doesn't have some of the drawbacks as v1.0.
v1.0 is still great for NSFW content imo, and I would stick with that version if that's the content you need to produce. For SFW and mild NSFW I would use v1.0 Alt.
The workflows and settings are the same for both versions.
I'm also working on a v2.0 that will combine the strengths of both versions, which unfortunately is not just as simple as doing a 50/50 merge between them.
FAQ
Comments (14)
every single example looks like overcooked, high-contrast, plastic "Flux" look. won't bother with this.
I don't agree that all the images have a plastic look but it's certainly fine if the style of the images I posted are not to your liking. However, that isn't the only style the model is capable of. This model is very sensitive to prompt direction, almost too sensitive. Like other Krea 2 models, it's capable of many different styles and just needs the appropriate prompt.
That said, v1.0 is far from perfect and I'm hoping v2.0 will be a better general use model. I'll try to include a greater variety of image styles in the showcase, although I thought I had done a decent job of that this time.
v1.0 ALT version is very good, can be used with character LoRAs without changing their character too much, great stuff!
One of the best checkpoints, the only thing is that it generates adult content poorly. It doesn't know poses well; it knows them, but the genitals and penetration are very poorly drawn. Also, the checkpoint doesn't understand the concept of an erection well. Hopefully, this will be fixed in the future. Thank you very much, I'll be waiting for updates.
Thanks for the feedback. It definitely needs some better training but at least we're not having to fight with it just to create basic nudity. I think Chroma is better at nsfw prompt adherence but Krea is better at detail. I may need to create a mashup workflow.
Hi @GBRX, I’m in my early generations of the ALT version and I can already say with confidence that it’s great!! The 1.0 version is interesting and a lot of fun to play with, but the ALT… great job. I can’t wait to see what comes next in this series—I think it could become one of your best series.
Thanks again and congratulations on the incredible work you do!
Thanks!
v1.0 seems to have MUCH better prompt adherence and understanding of sexual acts/nudity than v1.0 ALT, which sometimes just completely ignores the nudity, misunderstands details, and can be a bit difficult to work with.
For SFW generations, v1.0 ALT sometimes gives preferable results with regards to realism but not always; it's kinda case-by-case, but overall I like this version for SFW.
v1.0 ALT also seems to maintain more accurate likeness with character LORAs. I've only done limited testing of this but it's been consistently better than v1.0.
If v2.0 manages to be somewhat in the direction v1.0 ALT but with the adherence/NSFW capabilities of V1.0 then I think you'd have something really great.
I know this is pretty much in line with how you've described the models but thought I'd give feedback anyway.
Great work and thank you for all the effort you put in with your models. These are the first-and-only Krea 2 checkpoints I've used so far, and I've been thoroughly impressed!
Thanks, I really appreciate you took the time to test it this way and provide the feedback. I don't usually have much time to do character lora testing.
Personally, I like the original v1.0 over the ALT version because I think the aesthetics are better but it really sucks at SFW content so that needs to be fixed. ALT doesn't appear to be the full answer so I'm working on a v2.0 that I hope will keep the v1.0 aesthetics but work better for SFW too.
@GBRX No problem! Whether a checkpoint/LORA works well with character LORAs seems to be brought up a lot so I thought it was worth mentioning. It'd be great if v2.0 can keep this aspect of v1.0 ALT at least.
I'm not 100% certain which aesthetic I prefer more, but I will say that v1.0 ALT sometimes seems a bit more realistic overall, at least with regards to composition. And less noisy. Limited testing though, as I say.
I look forward to trying out v2.0!
which workflow is the best way to generate? Could you please share
All my images have an embedded workflow so use any of those.