<h3 id="i'm-back.">I'm back.</h3>realcum krea2 alphaThis is the first version for testing the concept with the krea2 model.triggers: cum on face,cum in mouth, cum covered face.This LoRA has been trained with a subset of my realcum dataset, about 15% just as a first test. Please have this in mind, my hardware is very limited and I cannot train with enough rank and the full dataset but still I think show great potential. I also planing to create a blowjob LoRA just because I think that concept is a complete mess right now and there is no consistent way to get decent results... Let's seeHave Fun!Currently I'm a little bit limited and I'm trying to get better training hardware, If you enjoy my work and want to help me support me at: <a target="_blank" rel="ugc" href="https://ko-fi.com/ale">https://ko-fi.com/ale</a> Thanks.v5 finalThat's it, I think this is the best I can do with this resources, I hope you all enjoy this model, the usage it's the same as v5.25q.Please leave a comment about what you like/dislike about this model and have fun!v5.25q!! Please read the docs and examples !!<ul><li>Changes<ul><li>Extended dataset: new images and lowres images removed.</li><li>New training workflow.</li><li>better details and prompt adherence.</li></ul></li><li>Useful Tags: cum on face,cum in mouth, cum covered.</li></ul>The rant:After a looong time and hundreds of training tests I have this new version, I think i'm hitting a wall in terms of what SDXL is capable of with LoRas. At least in my opinion this version is better than the previous one but just marginally better, maybe you even like the previous version more but it's up to you to decide.I really hope you like it because I saw sooo much porn trying to improve this lora that I'm now basically insensible to it.v4.5<ul><li>extended dataset</li><li>overall strength increased</li></ul>Hi again, I think I'm reaching the point of diminishing returns working on this, now it's hard to be sure that it's really improving or not, but I will keep working on this of out curiosity.Have fun!v3.1 no pp changes:Hi guys, this version removes all images with penises, I thought the penis concept will work fine but I was wrong, it was more annoying than helpful, so no pp this time.<ul><li>The dataset now is larger and higher resolution, you can appreciate it in the closeups.</li><li>Now it's also easier to prompt because the "solo" tag is no longer necessary.</li></ul>Usage:<ul><li>Useful tags: cum on face, cum in mouth, cum covered, (also tags like mouth, lips, alone can alter the result)</li><li>Models:<ul><li>Realistic models in general should work without issues.</li><li>NSFW models like sdxxxl works really well</li><li>acornIsBoningXL_v10 yields awesome results, it's a great combo, try it, it makes it 10x better.</li></ul></li><li>Please check the image examples so you can see how prompting works with different models!</li></ul>Post your results and have fun!v1 Usage:Models: I tested this lora with different models and works on almost anything (look at the examples) but I recommend Yamer's realistic (incredible results), Albedo base also works really well, even JuggernautXL works.Porn base models like Pyro's NSFW also benefit of using this lora.Negatives: This Lora produces nice results 1/4 times without any negatives but you can use it to make adjustment.Tags: keep the lora at strength 1, the tags can be use between 1 and 1.4<ul><li>main tags:<ul><li>cum on face</li><li>cum in mouth</li><li>solo</li></ul></li><li>secondary tags (not always works)<ul><li>facial</li><li>bukkake</li><li>covered in cum</li></ul></li></ul>Tips:- This model really like penises, the solo tag helps but if the base model doesn't have that concept can be annoying because adding penis to the negative doesn't help.- There is a tendency of showing the tongue, you can add "tongue" to the negative to mitigate it but the results depends on the base model.Advantages:- Works with almost anything realistic.- Easy to use.- realistic.- Extremely flexible, almost no impact in the original face.But Why?There is a lot of models replicating this concept, at least for me most of them doesn't produce good results or tend to produce similar faces.Also it's a great learning experience :)Leave some feedback and have fun!!

Real Cum

realcumk2.safetensors

realcum krea2 alpha

v6.55

Files with SHA256: ed26dd6c2c65b1b592498cc8ef720be202d4add3e779f087fc0b7758b6c95397