V0.9 - 57500 steps, v0.7 was 14500 steps, - Can pretty much generate pussy / penis via text prompt. - probably still need an additional lora for activation in sexual activity.
v1.0 will be around 75k steps or i feel like its done - then i will start a sex motion/action dataset and train it on top of this lora. and see what happens.
- T2V LORA - no real activation word but grok kept throwing "cel-shaded animation style" in there so it might help. note - no trigger words are trained.
str range 0.85-1
Creation tips
Remove Tiled decoding for normal vae decoding if your pc can handle it. much better
Reduce the secondary sigma values to reduce detail ruining. 0.65, 0.4250, 0.2219, 0.0 - seems to do ok
Enhanced motion - v1.0 | LTX Video LoRA | Civitai - lora for motion help - theres 100% better ones but this ones mine
BETA - Musubi keeps telling me its done and refusing to continue training - needs a slap - Will update in a few days
- Needs better cock training - i used PENIS LoRA by Coach Bate to enhance what the lora already knows but isn't good enough. .7 str
Sex actions probably need a lora - there's 1000's of sex action images but that's just to help activating everything TEXT TO VIDEO
cowgirl video - rogue-ai-missionary-cowgirl - was used .9 str
Total Visual overhaul - Better Motion - Less Visual distortion from a distance
So uhm, i spent 4 days on my 5090 laptop making 9000 images in Illustrious, - i saved every single caption along side each image This took just shy of 80 hours - the laptop did not turn off for the entire duration.
i don't even want to talk about the additional 15 hours it took to cache the text embeddings ._.
best video resolution is 1024x1536
issues -
Suffers from same face. - try to prompt different hair styles, might help shift it
didn't train enough cock -
Deffo more issues.
throw whatever you want at it - BUT IT WILL STILL NEED LORAS FOR MOTION ACTIVITES
im including my motion lora that should help movement with tits and stuff
BUT I BET - theres better motion loras on civitai.
fully automated dataset node i made myself to make every single prompt. 
(theres more images then then this list.)
