v2.5 of SexGod LTX 2.3 DoggyStyle Sex
version 2.5 should fix most if not all the noisy motion issues of previous versions. Put quite a bit of effort into this one, Have fun!
this is a full fine tune lora (meaning its adjusting all 22 billion parameters of the main model, so its big)
Im using a modified 19 step sigma schedule in phase 1 of my workflow (instead of 8 steps). Not sure if its better or not. The workflow below has the modifications I made to the sigmas, if you see Step 19/19 on your comfyUI console output vs 8 then its using it. Not required though, it will work with any LTX 2.3 workflow as far as I know.
Drag the image at the workflow URL below into ComfyUI to get the latest workflow
v2.0 Workflow: https://civarchive.com/images/127831837
Prompt example: "fast paced doggystyle sex, fully nude man dressed as a clown with strong arms and hips and a thich veiny erect penis is fucking a sexy young blonde woman hard from behind as the woman is on all fours, his penis is gliding in and out of the woman's vagina at a fast pace, nude woman with large natural breasts hanging and swaying as she leans over, her large natural breasts and erect nipples bouncing and swinging with every powerful fast thrust of the man's hips, the man thrusts his thick veiny erect cock in rhythmic motion, the man is fucking the woman fast with rhythmic thrusting of his hips, the woman says in a moaning orgasm filled whispers "Oh Ronald! Your secret sauce is the best sweetie!", her mouth open in pleasure, realistic skin texture, detailed breast and nipple movement, no motion blur, indoor lighting"
If you are using image to video, try the Klein 9b couples nudity lora below to get the subjects in the right position first.
Description
FAQ
Comments (44)
I am downloading and will rename it to "humping Ltx2.3" lora...just sayin'
not exactly sure what you mean?
@sexgod1979 Honestly, most people can't read. Like what you've done with these examples is more impressive than what these people expect. If they read the prompts, they'd see it's doing exactly what is specified. It's on them to become literate. We all know America isn't just post-truth, we're post-literacy.
@happylittleteapot I just started posting some videos, takes a while to get various I2V and T2V poses all rendered. I do my best, its not easy making loras for LTX 2.3, but I try my best to support the community with something fun we can all enjoy
I ain't never seen a dog do it like that.
no workflow found in video? would you mind upload a png with the workflow?
yeah I dont get why when I have save meta data the node doesnt actually save the meta data. I just posted a screenshot of the workflow. If you look at my workflow image and can tell me why its not saving the meta data, im all ears!
ask AI and get this, maybe just the meta data too big to fit in mp4 , or the encoder issue, PNG seems always works.
✅ Why PNG Always Saves Metadata
PNG files support embedded metadata very easily (tEXt / iTXt chunks).
ComfyUI writes the full workflow, prompt, seeds, settings, etc., directly into the PNG file.
PNG is basically “metadata-friendly by design.”
That’s why:
You can drag a PNG back into ComfyUI
It reloads the full workflow
All settings come back perfectly
❌ Why MP4 Often Does NOT Save Metadata
MP4 is a video container format, and:
It does not natively support arbitrary large metadata blocks like PNG.
Many video encoders (FFmpeg, H.264, etc.) strip custom metadata.
Some players remove unknown metadata fields during processing.
Some ComfyUI video nodes simply don’t write metadata at all.
So even if ComfyUI tries to save metadata:
The encoder may ignore it.
The container may drop it.
The metadata field may be limited in size.
Or the node you’re using doesn’t support it.
@unheaded hrmm okay. I added the screenshot of my workflow and included workflow link in resources section (all I did really is add a power lora loader from rgthree)
@sexgod1979 if you use the VHS node to save the video, the audio version does not save metadata, the one without the audio does, its a weird issue. If you use the native comfyui create video nodes then this is not an issue for the audio version of the file since it only makes one output only.
@Buzzard19 oh wow ok. but I love the audio!
Very nice💕
2,5gb its a joke.
for a rank 64 lora on 22 billion param model, I assure its pretty much the way it goes. And what does it matter, just let comfyUI swap. Doesnt cause any issue for me. I can load the entire 42GB gguf bf16 model no problem
lTX 2.3 even in fp8 is a huge model, I think the bare minimum to run it is 24GB of vram, comfyUI is good at block swapping
kjNodes has a LoraReduceRank based on the kohya-ss script, meaning you could use it as well. You can always reduce the size, just know you may lose some of the definition of the original rank. You could also attempt to remove certain layers if you know they aren't useful, but that's even more advanced than I know lol.
It's not a big deal if you have VRAM+RAM, but can understand if your Harddrive is filling up lol
@happylittleteapot Go for it if you like. Many of the LTX 2.3 I look at are this size, not all depends on the training rank and what attn block they targeted but many are 1.5+ GB
something i'm testing at this moment is training first at 64 reducing at 32 at 2000 steps , one of the best lora on ltx2 is rank 24
@JaimyB yeah im not sure if its due to it being rank 24, rank 24 is quite low. Likely other factors, but honestly I dont know. LTX 2.3 is a bit of a pain in the ass to be honest to train
@sexgod1979 so true..
@JaimyB One of the best LoRAs recently released had its rank shrank after training. Never assume a LoRA was actually trained on the rank you see.
@huj0ps1t6 thats interesting thanks
pls do a handjob lora
no good ones?
@sexgod1979 I could only find Penile Praxis and Dreamlay but there isn't a specialized one yet
@akira Okay cool, ill see what I can do then :)
your wish is my command, https://civitai.com/models/2542266/sexgod-handjobs-ltx-23?modelVersionId=2857106
I don’t know see any doggy stylists in examples? Is it April fool?
really? what position would you call that
@sexgod1979 first few are standing from behind, oh last few are last few are doggy style, bent on knee/ kneeling
@iluvlamia we call that 'standing doggystyle' where I am from :)
@iluvlamia but its capable of many doggystyle poses, im just being lazy with my videos
Where is the link to your workflow included? I can't seem to find it.
Facechanger or not?
its been holding the face for me with image 2 video using the workflow in the description.
however if you are applying a character lora on top in text to video mode, I cannot honestly say as I have not tested that
Actual sex god; works awesome !
sweet!!
For me, in i2V wf, nipples poke through clothes.
Any idea how to resolve this issue? Happens only with this lora. Tested with str .4, .6, .8, 1. On .2 works correctly, it appears.
EDIT:
It seams like this lora just bump up power of other loras, that I used alongside, and those do cause this issue.
Seams to be working very well.
Thank You for this lora !
This one is awesome! Just what i needed!
Try version 2.5
2.5 gig?
yeah, I tried to fine tune the entire ltx 2.3 model which made the lora huge. I am going to work on reducing the file size. Sorry about that