🔗 Reverse Suspended Congress LoRA (Video | WAN2.2 5B)
Hey everyone!
This is my fourth video LoRA, trained for the reverse suspended congress position — an intense and dynamic sex pose involving full-body suspension and penetration from behind. Built using WAN2.2 5B, with Ostris’ AI Toolkit and the official training tutorial — same simple process!
This LoRA isn't that good probably need a better dataset and training.
🔧 Training Details:
Model Target: WAN2.2 5B
Dataset: AI-generated 3-second clips (72 frames), built with WAN2.1
Focused on reverse suspended congress position, showing strong hip movement, leg suspension, and penetration angles
Angles Included: Frontal, and partial over-the-shoulder
Resolution: 512x512
Training Steps: 3000 total
Best Results: Around step 2500
Training Time: ~4 hours on an RTX 5090
VRAM Usage: Max around 14GB
Workflow: Default ComfyUI video LoRA setup with LoRA loader
Trigger Word:
RSC
💡 Notes:
Dataset emphasizes motion, weight suspension, and intimate body contact
Could be improved with better leg physics and more consistent body alignment
WAN2.2 5B is fast and light — works well despite lower output quality than 14B
Hope this helps creators working with extreme poses or looking to expand their LoRA catalog!
Always open to feedback or training tips 🎬🖤💦
Description
FAQ
Comments (12)
Have you had any success training character loras for 5B? I trained one a couple of days ago and I'm not sure if I'm satisfied.
Nope I only trained NSFW concept so far.
I think it breaks TOS. Maybe check out a place called datadrones. Maybe search for Sasha and maybe filter for 5B loras. Maybe.
wow. is this 5B? amazing!
5b cumshot?
Any plans to do 14b? Would you share your dataset on the post?
No plans yet but probably soon when ai toolkit support it, the dataset was generated videos using wan 2.1
it tends to make the girl fly like a rocket if you goes after up and down xd. but rest is cool
yo can anyone give any basic work flow for this?
one for solo fingering and masturbation maybe?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.