This model is for creating inverted fellatio image generations using the side view. It is not trained on POV inverted fellatio whether the male or the taker POV for that matter.<h2 id="about-this-version"> About this version</h2>This model has a strong 3DCG bias. Much of the training data came from generated images from my private LoRA, which was overtrained on 3DCG and one particular catgirl video. Some of these biases have leaked in through using too many stills from that video in this version. Expect a revision of this LoRA in the future.Also important: This LoRA tends to make a double-decker bed so to avoid this, prompt the floor i.e. tile floor, wooden floor. Example prompt: 1girl, solo focus, inverted oral, sloppy bj, on bed, on back, long hair, head hanging over edge, head tilted back, fellatio, balls deep, closed eyes, milf,Prompting 1boy is not required in most cases. If you attempt to use this with realism checkpoints, increase the strength to 1.1 and the CFG to &gt;7.0., and prompt 1boy. But due to the high LoRA strength, expect a freakshow. You may be able with cropping to obtain a useable section of the image (that's how I plan to obtain more realistic training data for the revision, but anyhoo).<h3 id="on-angle">On Angle</h3>Using both side view and medium angle in the same prompt is likely to confuse the AI. Medium angle and side view are likewise meant to be exclusive, however the AI tends to ignore them so try emphasizing them.Far more important than prompting angle is giving the model permission to adhere to the close-up prompt by omitting legs/feet from the prompt entirely. If you prompt her feet/shoes/thighhighs, you are going to wind up with those elements in the picture and likely her legs up in the air in an unnatural pose. <h3 id="a-word-on-recommended-models">A Word on Recommended Models</h3>First of all, I could not have created this model without Over the Edge by @freckledvixon and that model is trained on a more varied set of data and can handle a lot of angles. Definitely a must-have in your arsenal.As for this model, due to the bias in the training data, you may notice more training bleedthrough such as pink and blue neon signs plus rain (if you prompt "night") or black cat ears on the girl. You may find this bleedthrough suddenly appears when switching from a cartoon or realism LoRA to something like Nova 3DCG XL and suddenly the LoRA isn't fighting the model so much. In these cases lower the strength of the LoRA.All example images were generated only with the LoRA and the checkpoint, 28-35 steps, 4-5 CFG.

Inverted Oral Side View

This model has a strong 3DCG bias. It was not a merge with Over the Edge by Freckledvixon and if you cannot achieve the results you are looking for with this model I HIGHLY recommend that LoRA.

Files with SHA256: f95bd4b0045b122d03bd00922b4f3f821c12d1e0bba2f9bc6818ca3d97765d71