A motion-focused LoRA trained on 40 HD videos to reduce stutter, improve pose transitions, and add micro-movement “life” to NSFW shots. Works great in ComfyUI video pipelines and can add dynamic energy to stills. You supply the prompts; this LoRA supplies the fluidity.
What it does
Smoother motion between frames (less jitter, fewer frozen frames).
More natural body sway, hair movement, and camera drift.
Better temporal consistency for skin, fabric, and highlights.
Plays nicely with realistic and stylized NSFW bases.
Requirements & compatibility
Video: ComfyUI (tested pipeline recommended).
Install
Drop the
.safetensorsinto:ComfyUI: use a Load LoRA node and connect to your base checkpoint.
Quick start (stills)
Negatives (baseline):
bad hands, extra digits, deformed, text, watermark, overexposed, oversharpen, lowres
Prompts
A woman straddles a man during sexual intercourse, performing rhythmic movements as she bounces up and down atop him.
Training details
Dataset: 40 carefully curated HD videos (varied poses, angles, and lighting).
Focus: bodily micro-movement, hair/clothing dynamics, camera drift.
Preprocessing: high-quality frame extraction, deduplication, face/skin priority.
Objective: Temporal coherence + natural motion cues for NSFW scenes.
Known limits
Extreme low-light or strobe scenes can reintroduce shimmer.
Very aggressive sharpening can create “crawl”; keep denoise moderate.
Changelog
v1.0 — Initial release: 40-video motion LoRA, tuned for ComfyUI.
Description
wan2.2_t2v_highnoise_betternsfw_v1.0
FAQ
Comments (7)
Looks great!! But could you post some preview videos?
This would be good as a High Noise lora for Wan 2.2, since high noise is where most of the movement happens.
Any samples with that "good motion"?
The file names say T2V but the description base model says I2V?
I think your Wan 2.2 lora is an upgrade.
I'll reserve judgement until I give it a go myself, but the example videos don't really inspire confidence. Movements look pretty robotic and stilted. Granted, all the examples so far are from the uploader, so it might just be something on their end rather than a fault with the model itself.
I'll give some testing on my end to see how it stacks up against the other generalized NSFW lora. Who knows, maybe they might work well together.
File name says T2V but its posted under wan i2v. Which one is correct?