I am glad people are having success with this embedding! I am releasing an updated version that should result in more realistic results.
Uber Realistic Porn Merge (URPM) | Civitai gives great results with this for realism and inpainting!
Negative prompts can have a large effect on this embedding, both suppressing it but also possibly increasing the quality. I use the 'bad prompt 2' embedding at a weight of 0.3, found on Huggingface here: Nerfgun3/bad_prompt at main (huggingface.co).
'Restore faces' should be off for best results for closeups but can help when the subject is further away.
==
This embedding is used for adding a bukkake effect to generations. It can also be used for inpainting will good effect, see attached screenshot for v1 (denoise is 0.5 there.) Quality varies per model so experimentation may be needed.
I generally prompt in the format 'blah, blah, a picture of bukkakAI woman, blah, blah,' but it should be flexible. For inpainting the embedding name alone should be sufficient.
This was created based on images taken from ImageFap galleries.
Description
Expanded dataset to 188 images, 0.005:8,0.0005:140,0.0001, 176 steps, 37batch/4grad.
FAQ
Comments (20)
Just a thought can someone with experience in creating embeddings or checkpoints try to train a model with high-res pictures 2048px or more of hands, fingers and eyes to solve that problem of bad mutated hands and ugly eyes once and for all? I think we have now enough models of vaginas, tits and dicks. 😂😂😂 LOL
You can try bad-artist and bad-prompt negative embeddings to get better hands.
can this be updated to version 2.0 ore higher?
Hi, if you're talking about using this with SD 2.x instead of 1.5, I'm afraid that's not in the cards yet but hopefully in the future unless someone else beats me to it.
If you're just talking about the v2 of the embedding, just replace the original .pt file with the newer one.
@psm3
I was talking about the SD 2.x.
tank you for the replay.
For version 2 note the trigger word is bukkakAI2
Heya, it is whatever the name of the embedding is, I kept it the same for each version but you can make it whatever you want. There's no trigger other than that.
Could someone please explain how to use this embedding? I added it to the embedding folder, added "bukkakAI2" to the prompt, and used it with the Uber Realistic Porn Merge model, but it's not producing any results. :/
Check the name of the embedding. In the console it should say something like: "Textual inversion embeddings loaded: bukkakAI.
I'm guessing you just have to remove the 2 since version 2 is called bukkakAI by default as well
I added it to InvokeAI and it shows up as "dec22test2" when Invoke scans the available embeddings. Try using that as your trigger phrase.
It's just not working for me with URPM inpaint. embeddings loaded: bukkakAI
make it safetensors plz
Heya it's just an embedding, safetensors is only for checkpoint files :)
i am using the embedding as instructed, but so far I am getting nothing? is there any extra steps besides putting it into the embedding folder? i have tried putting the trigger into both the positive and negative prompts, I've even tried a couple of the things I have read on here.
Are you doing inpainting or txt2img? If the former double check denoise and other inpaint settings... the latter, if it doesn't do anything with a very simple prompt like 'a photo of a bukkakAI woman' then I'm not sure, sounds like a more fundamental problem...
@psm3 You should put that prompt in the description. I was struggling too until I used it. Great work btw. :D (ya I didn't click the MORE button, I see it now haha)
Can somebody let me know how to add this to Draw Things? Or is that not possible? Thanks!!
i don't get white stuff when i inpaint, i get colored blobs that look like slugs
For anyone struggling like I was for a VERY long time to get Inpainting working, these settings have been pretty reliable to get better output. Any suggested changes are welcome though!
Prompt: bukkakAI
Model: uberRealisticPornMerge_urpmv12.safetensors
Resize Mode: Just Resize
Mask Blur: 4
Mask Mode: Inpaint Masked
Masked Content: Original
Inpaint Area: Whole Picture
Sampling Method: Euler A
Sampling Steps: 30
CFG Scale: 7
Denoising Strength: 0.5
Seed: 3394251902
All other options are set to default. I took the Seed from the example images but if you find one that works for you then stick with it.
I recommend building up the image one "spurt" at a time and re-send to Inpaint, rather than masking off all images of the face you want to coat in buttermilk
I have a question. It doesn't seem to apply to fresh text2img generations for me, but works fine with img2img. any suggestions on how to make it work with new text2img generations?