Ever wanted to know what a cervix looks like? Ever wondered what it would be like to see a character inside a female appendage? Look no further than this model! Vore? You got it (sans the teeth, just the insides)! Unbirthing? You got it! Caves with walls of flesh like the ones frequently depicted in Sci-Fi, anime and some of those tacky 1980s shows... (admittedly, I liked them too...)? You got it!
This model is very good at depicting the insides of a person. The only issue? It has a slight contrast problem... So it pairs well with some of your upscalers (which add too much contrast). Come download this model today.
$5.99 +S&H while supplies last...
(Not really, the model is free to download.)
This model works a little differently than other LoRAs.
⮊ The strength indicator is going to strongly influence how the end image turns out.
⯄ Use low strengths (0.1-0.3) for when you want bodily insides (vagina, anus, arteries, etc.) to look a bit more... good looking.
⯄ Use mid strengths (0.4-0.7) for when you want to add fleshy walls and/or want a more pink-looking image
⯄ Use high strengths (between 0.8-0.9) if the model is uncooperative with the LoRA. You can also use "the ring" trigger word with this (or any lower strength) to make the LoRA do its charm.
⯄ Use ultra high strengths (0.9-1) if you want to produce an image of a cervix or another internal organ. At these strengths the model is disobedient. It might produce what you want, but it is more likely that it does not actually generate what you want it to generate.
⮊ This model does not need the trigger words. Most of the time it works without the trigger words. Add the trigger words for extra strength if needed (or if you want the image to include an all-seeing cervix).
⯄ P.S.: I strongly suggest you use "in between" and "non-conventional" numbers (like, lets say... 0.874, or 0.626 or 0.543). Use any number in between that will yield the desired image. The key is in the number and the strategic use of the trigger words "the ring".
Description
This model works a little differently than other LoRAs.
⮊ The strength indicator is going to strongly influence how the end image turns out.
⯄ Use low strengths (0.1-0.3) for when you want bodily insides (vagina, anus, arteries, etc.) to look a bit more... good looking.
⯄ Use mid strengths (0.4-0.7) for when you want to add fleshy walls and/or want a more pink-looking image
⯄ Use high strengths (between 0.8-0.9) if the model is uncooperative with the LoRA. You can also use "the ring" trigger word with this (or any lower strength) to make the LoRA do its charm.
⯄ Use ultra high strengths (0.9-1) if you want to produce an image of a cervix or another internal organ. At these strengths the model is disobedient. It might produce what you want, but it is more likely that it does not actually generate what you want it to generate.
⮊ This model does not need the trigger words. Most of the time it works without the trigger words. Add the trigger words for extra strength if needed (or if you want the image to include an all-seeing cervix).
⯄ P.S.: I strongly suggest you use "in between" and "non-conventional" numbers (like, lets say... 0.874, or 0.626 or 0.543). Use any number in between that will yield the desired image. The key is in the number and the strategic use of the trigger words "the ring".
FAQ
Comments (5)
I recommend combining "the ring" and "the source" models for more accurate "fleshy walls".
Any chance to train on higher resolution and with reduced network dim? Dim 256 is quite high and thus the LoRA size is huge. 320x576 is quite low for the resolution. Have you used anime images or real pictures?
You can generate images at high resolutions. I recommend 512x768 or 768x512 and then using HiResFix. I can try to reduce the LoRA rank to something smaller, but there is a possibility that the LoRA won't behave the same way as it does with this rank size (I'll upload when I have a chance). I used real images to train the model, so (in theory) you can generate real-life style images. You can generate animé images using an animé model. I did that using DucHaiten's God of SIMP model (currently shown in the gallery below), but you can use any model that can generate images in animé style. Finally the 320x576 resolution dimensions were chosen because the average size of images were roughly around that size, so if I increase the resolution the output images generated with this LoRA won't look crisp and will have severe contrast issues. Upscaling the images introduced artifacts which is why I chose to train the LoRA in their native size. Maybe if I can source good quality images at larger sizes: I might be able to train a LoRA at a higher resolution. Does this answer all your questions? Feel free to ask again if I didn't. Also feel free to ask new questions if you'd like.
I created a resized version of the LoRA. It behaves a little differently, but still produces great images (I don't really know what you're using it for, but what I'm using it for: it produces good stuff). I hope you like it. Check the LoRA page for the model (https://civitai.com/models/561494/the-ring).
Yea, training resolution and generation resolution should match to get best results. Also its best to use the same model for training and generation. My close up LoRA's almost only work well on my model and on other models it look bad. So if you train real images on an anime model you wont get the best results if you want to use a real model for generation afterwards.