[EDIT_260110:]
Alright folks, here you go — the high-noise model is now available as well. Since I got pretty heavily attacked in the comments, that stuck with me for a while, so I didn’t exactly rush to make this public…
The workflows for the images shown are available on my Patreon along with more Training Epochs. Please support my work.
Have fun with it. <3
Frank
https://www.patreon.com/polyhedron_ai
-----
So, folks, here’s the body-fix for nude women as a finished low-noise model, the one I already made for FLUX… Unlike that, the WAN 2.2 model actually understands pretty well what nice breasts and nipples are. :D But anyway, I trained this one too—maybe it helps bring a bit more realism into images and videos and counter that plastic look you get from accelerators like FusionX or Lightx2V.
The images shown were also made with both accelerators; they went through a high-noise path and a low-noise path…
On my Patreon page you’ll find the full workflow for this, plus the high-noise model with different training stages.
Hint: If you only want to use this low-noise model—which definitely has its advantages—add it as a LoRA in the high-noise path with a strength of 0.6. If you use the high-noise LoRA instead, lower that strength to about 0.35. You can also use my “Polyhedron_all: Perfect skin. Perfect hands. Perfect eyes. (m/f)” at the same time:
https://civarchive.com/models/1947995/wan22-t2i-highlow-noise-polyhedronall-perfect-skin-perfect-hands-perfect-eyes-mf
Just give it a try. Experiment. Experiment. See what works for you. <3
Frank
P.S.: The high-noise model will also be published on Civitai, but at a later time.
Description
V.10
FAQ
Comments (12)
Maybe my brain is slow today but this sentence is so confusing to me right now: "If you only want to use this low-noise model—which definitely has its advantages—add it as a LoRA in the high-noise path with a strength of 0.6. If you use the high-noise LoRA instead, lower that strength to about 0.35."
Can you help my slow brain? What do you mean if you use the high noise LoRA instead? You only uploaded a low noise LoRA?
Low Noise: Here! (Use it also in the high noise-path with a strength of 0.6)
High Noise: Patreon. ^^ (If using this, reduce the strength to 0.35)
@Polyhedron_AI It all makes sense now, it seems it was indeed my slow brain :D
Thanks a lot, will try it bc i love your other Lora, made tons of pics and videos with it!
@Lora_Addict You're welcome. ^^ Enjoy.<3
Marketing is confusing on a website where the models are free.
@p_huc_h8erz_kek YEAH! This website created all the datasets for my LoRAs, trained everything on paid high-end machines on Runpod, and handled the entire workflow in ComfyUI…
No, wait: that was me! ;)
@Polyhedron_AI - your comment is very confusing. It is super unclear what you are trying to say, yet again.
@p_huc_h8erz_kek
Wan 2.2 uses two models for text-to-video: a high-noise model and a low-noise model.
Therefore, in ComfyUI you need two generation paths: first the high-noise path, then the low-noise path.
You can train LoRAs for both models:
– LoRA 1 for the high-noise path
– LoRA 2 for the low-noise path
If only the low-noise LoRA (LoRA 2) is available, you can still use it in the high-noise path — with a higher strength of about 0.6.
If you also have the high-noise LoRA (LoRA 1) — which is available on my Patreon — then use LoRA 1 in the high-noise path instead, with a lower strength of about 3.5.
Clearer?
P.S.: The high-noise model will also be published on Civitai, but at a later time.
@Polyhedron_AI - absolutely nothing about any of this is actually unclear you smarmy frip. You are withholding for profit. Offering literally half of the product for free and dangling the other half for pay.
Your mangled english is a lame and failed attempt to market your wares without offending people's sensibilities.
This is disgusting behavior and you know it.
And if you were smarter maybe you would use a training script that allows you to train a single file instead of two? lolol
the gall of you trying to preach to me about how training LoRAs works when you yourself know barely anything.
maximum kek
shiesty af fr fr
lrn 2 mrkt bro
@p_huc_h8erz_kek Yes, I share my models, do all the work, and still get insulted for it. Great. Try telling your local barista that your cream latte with extra sparkle should be free. If you’re unhappy with my work, use different models—there are plenty out there.
@Polyhedron_AI - top kek
your takeaway should be
"man maybe I need to figure out how to market my wares without being a shady shiesty POS... perhaps I could mimic the effective and respectful creators on civit who don't insult the userbase..."
not
"wow so mean bro wow, ok, wow"
You know? Life is hard enough without you whinging about being mistreated when you are a lowlife bait-and-switch charlatan...
You are being intentionally deceptive.
You are using obscure language that is literally confusing people because you know that what you are doing is wrong.
Most creators monetize, but no other creators use bait and switch techniques and use intentionally obfuscated language to deceive people.
You are not being insulted for providing free models.
You are being rightfully called out for being underhanded and shady.
FWIW I didn't download your weakass model bro, if I thought I needed some better LoRA for nips than already exists... and there are sooo many very nice and effective ones already... IF I thought I needed one, I would absolutely train it.
Too busy training fun and interesting models for free... no need for more bloated overtrained weak boring nude body LoRAs fren, but thanks anyway.
@p_huc_h8erz_kek Sure, as you wish... :) Keep it going!
Details
Files
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
Mirrors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
polyhedron_wan2.2_nudes_low_noise_low_noise.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.