Update: Reverse deepthroat for LTXV2.3
Trained on LTXV 2.3. Improved captioning and way more training steps at full frame rate.
Keyword: reverse deepthroat video.
Try natural prompts ... for example:
ballsdeep_throatfuck, he fucks her throat balls deep
huge throatbulge, her throat bulges (exaggerated and prominent) as his penis as he thrusts his penis into her throat
his balls slapping against her nose
dynamic_throatfuck, he fucks her throat penis in and out,
he stands on top of her. he inserts his penis into her mouth
pushes in against throatbarrier
... whatever you want to happen.
Guarantee: the training set did not contain ball on top/in front of penis ... and still this seems to be an issue with LTXV generations. Negative prompting (balls/testicles) might help.
Reverse deepthroat aka upside down face fuck LoRA for SDXL, PDXL and WAN.
Keywords for image generation:
reverse deepthroat, kneeling, sitting, leaning back, arm support, lying on back, supine, balls deep, penis pulled out, neck grab, breast grab, holding head, holding down, no hands, side view, frontal view, closeup, nipple pinch, breast squeeze, lips that grip
Major update for WAN2.2. Way more training data. Twice the steps. POV and male appearing should now work.
Keywords for video generation:
"reverse deepthroat" + optional combinations of "oral insertion","deep throat fuck", "dynamic throat fuck", "male appears", "in and out", "throatbulge".
Description
FAQ
Comments (3)
Well for some reason I’m getting a lot more lucky with this lora, it’s not exactly what the prompts in the examples show but it’s pretty good! Thanks for making it
Mind posting a few? Which one is it you are using ... WAN? ZIT?
@LFMX I'm using the standard Text 2 Video Wan workflow from comfyUI browser I believe and I used the prompt from one of the examples below:
Caucasian male, her hands at her sides, ( reverse deepthroat:1.3), fellatio, deepthroat, cock suck, dick suck, blowjob, lying on a sofa, penis, (((neck bent way back over the edge of sofa))), (((((woman facing up))))), ((((close up of face)))),
Then I had the high and low at 1.0 and 9 out of 10 worked without a problem.





