<h2 id="to-the-mod-teamtos-disclaimer">To the mod team/TOS disclaimer</h2><p>This model complies with the TOS to the best of my knowledge. For it to operate the person must already be in a sexual scenario(eg a woman and a penis must be in the scene). It, by itself, does not add a penis into the scene. None of the training material included anything that took a person and moved them into a sexual scenario, it all started with both man and woman in the scene. Happy to show the training material to the mod team if that helps at all.</p><p></p><h2 id="usage">Usage</h2><p>Ok now that is out the way, lets get to the model. It's basically a blowjob lora but where the man has a flaccid penis. As mentioned you'll need a man and woman in the scene at the start for it to work. I've tried to train it on a variety of shapes/sizes. So base prompt is:</p><p></p><pre><code>A woman gives a man a blowjob. </code></pre><p>followed by some description of the penis. Your options are:</p><pre><code>The man has a large flaccid circumcised penis. </code></pre><pre><code>The man has a large flaccid uncircumcised penis. </code></pre><pre><code>The man has a small flaccid circumcised penis. </code></pre><pre><code>The man has a small flaccid uncircumcised penis. </code></pre><p>Finally followed by the angle of the camera, two options here:</p><pre><code>The camera angle is a side view.</code></pre><pre><code>The camera angle is from the man's POV.</code></pre><p>That should be enough but adding how she gets from where she is to the penis never hurts.</p><p>I'd suggest starting with a 0.5 weight for both high and low and then upping it if you aren't getting the desired effect. Sampler wise most should work but i'd avoid unipc as it does some weird things. Stick with Euler/LCM/Res.</p><p></p><h2 id="training">Training</h2><p>I don't know if anyone finds the training details helpful but maybe someone can glean some value from them. The model was trained on 77 videos all 5 seconds long and 16 fps. High noise was trained at 288 resolution with 4 frame buckets. Low noise was 608 resolution with 2 frame buckets. 50 Epochs at 4 batch for High and 2 batch for low. All trained on diffusion pipe on 3x3090s and 1x5060ti.</p><p></p><p>Captioning was automated using JoyCaption and DarkAges Llama 70B.</p><p></p><p>Enjoy the pp physics! </p>

Files with SHA256: b4a6e75c60876463a64ffb50955fc5d777ca5145890b4a8ec7450fba1948e3e8