EDIT: (260509)

Okay, you beautiful maniacs — here's the High-Noise version as well … And as usual, I've continued to evolve my work over the past weeks; the images shown don't represent the LoRA alone, but rather the current state of things: I've used several new LoRAs here that crack the base model of WAN 2.2 open …

I've also used a brand-new custom node that I vibe-coded with Claude over countless hours: "Polyhedron LoRA Stack" — it not only has a ton of great features on the frontend (per-LoRA thumbnails, trigger word management, a canvas UI), but really shines on the backend: group-based LoRA merging in DARE, CONCAT, and more — not just the sequential stacking you know from the Power LoRA Loader.

You'll find everything you need over on my Patreon: all the LoRAs used, the current workflow, the sample images with metadata, and of course the custom node(s). Enjoy!

And yes … I'm still on stills. What I mean by that: how everything behaves for actual video generation with multiple (video) LoRAs — I honestly have no experience with that yet. Videos are coming once I'm happy with them … Stay tuned!

https://www.patreon.com/c/polyhedron_ai

------------------------------------------------
Alright, you crazy rocket scientists,

this project has been driving me completely insane over the past few days — thanks a lot for that. :D I had created a nice LoRA for “rain” and wanted to properly finalize and share it. And then, being the naive person that I am ^^, I thought: “Hey, how about a nice waifu as an example, casually strolling through an alley in the rain — soaking wet, dripping, with sexy tits: sounds doable, right?” Wrong! :P

I’ve run into this issue before with FLUX and generally with earlier AI models: it still seems difficult for a model to convincingly render wet fabric on wet skin. So I went ahead and trained another LoRA for “wet shirts,” which I’m sharing with you here — for now as a LOW NOISE version.

The HIGH NOISE version, as usual, is available on my Patreon page; there you’ll also find the images, my current ComfyUI workflows, and the new “rain” LoRA, which I plan to share here at a later point as well…
https://www.patreon.com/polyhedron_ai

For training, I only used example images(!), and I think it turned out okay — but it really shines when you mix in two external video LoRAs that I’d like to point you to:

Braless Walking Wan2.2-T2V-14b by Fabin_2024:
https://civarchive.com/models/2016969/braless-walking-wan22-t2v-14b-and-wan22-ti2v-5b

Wet Translucent Clothes (Wan t2v) by CaptainWebSpinner:
https://civarchive.com/models/1722244/wet-translucent-clothes-wan-t2v

In general, I also threw in the LoRAs I had already shared before… and the final results look a lot better. Yeah.

Give it a try! Right now I’m only doing single images because I want to do my homework first — learn to walk before I try to fly… but hey, maybe videos are fun to look at too. Feel free to show me what you come up with. It’s still not perfect — the eyes could be better, the hair could be better, the nails should look much nicer after upscaling… but what is perfect, anyway? I need to let go. mic drop

Frank

P.S.: Still include this LOW NOISE model in the HIGH NOISE path. The other model will be released on CIVITAI in the coming weeks — no worries. :)