CivArchive

    Support me so I can make more models, faster: https://ko-fi.com/the_cook
    The aim of this model is to help create nsfw "nude portraits", with the person standing and posing like they would at a photoshoot. It's also supposed to help with the physics and structure of anatomy. The dataset consisted of about 220 captioned videos, with varied faces so that there is minimal influence on your character loras. I tried to minimize the occurrence of jewelry, piercings, tattoos, long nails, and captioned accordingly, so you can mention any of these in the negative prompt to get rid of them.

    The training was done at a 2:3 aspect ratio, so you can use these resolutions:
    256x384, 384x576, 512x768, 640x960, 768x1152, 896x1344, 1024x1536

    Currently this lora is best used with these distillation Loras:

    https://huggingface.co/wangkanai/wan21-lightx2v-i2v-14b-480p/blob/main/loras/wan/wan21-lightx2v-i2v-14b-480p-cfg-step-distill-rank256-bf16.safetensors on the Low Noise at 1 strength.

    https://huggingface.co/Kijai/WanVideo_comfy/blob/main/LoRAs/Wan22_Lightx2v/Wan_2_2_I2V_A14B_HIGH_lightx2v_4step_lora_v1030_rank_64_bf16.safetensors on the High Noise at 0.5 strength

    Euler / beta or beta57 works best with these distillation loras, 8 steps split between High and Low samplers

    You can try this as your starting prompt: "A beautiful nude woman at a photo studio photoshoot, she is posing seductively and then turns around and looks back. She is completely naked. Professional studio lighting, high quality production"

    You should be able to control various actions with prompting, like hand placement, body turning etc.

    Also try this prompt with V2: "The clothing suddenly disappears, revealing her naked body. The woman is now nude, and she poses sexily."

    Description

    More training, better data, more capabilities (see examples for what it can do. also take a look at the embedded workflow)

    FAQ

    Comments (66)

    WearCatEarsNov 10, 2025· 4 reactions
    CivitAI

    Probably one of the best I2V WAN 2.2 lora to unlock nudity for any body shape.

    1601658Nov 10, 2025

    @WearCatEars Hi, do you use Windows or Linux?

    WearCatEarsNov 10, 2025

    @nokiasupreme973 Windows, why ?

    1601658Nov 10, 2025

    @WearCatEars Thanks for replying. I'm using Windows and I keep having problems with the ComfyUI workflows for WAN 2.2.

    The second environment I created for this (ComfyUI Portable) just broke, either because of faulty nodes I downloaded or because I tried to run Triton on Windows.

    1601658Nov 10, 2025

    @WearCatEars I can't find any stable workflow.

    WearCatEarsNov 10, 2025· 1 reaction

    @nokiasupreme973 try this portable version of Comfy https://huggingface.co/ClawCraft3r/ComfyUI/blob/main/ComfyUI_windows_portable_3.62_SageAttention2.2.7z sage attention and triton already enabled + custom node for wan 2.2

    1601658Nov 10, 2025

    @WearCatEars my god, I'm going to try what you shared. Thanks!!

    1601658Nov 10, 2025

    @WearCatEars I downloaded it. When I can run, I obtain this error message:

    No input found for flattened id [145] slot [-1] 

    WearCatEarsNov 10, 2025

    @nokiasupreme973 which workflow are you using ?

    1601658Nov 10, 2025

    @WearCatEars WAN2.2_T2V_I2V_ALLINONE.json, of https://huggingface.co/ClawCraft3r/ComfyUI/tree/main

    WearCatEarsNov 10, 2025

    @nokiasupreme973 You need to go to the subgraph (double click on them) linked to the models and change them to the models you have. You also need to use GGUF versions for this workflow. Then don't forget to choose the mode you want to use in the node with the switches. Otherwise, I can upload a simpler version of the workflow if you only want I2V.

    1601658Nov 10, 2025

    @WearCatEars Thank you for your time, but I can't use FP8 or lower models because I have an RTX 3090. I can only use FP16 and above.

    WearCatEarsNov 10, 2025

    @nokiasupreme973 You still can use GGUF model, try Q6 or Q8, it's going to be fine. https://huggingface.co/QuantStack/Wan2.2-I2V-A14B-GGUF/tree/main

    1601658Nov 18, 2025

    @WearCatEars Hi, I've been testing your workflow for several weeks, and it's finally working for me. Since I only use FP16 models, I removed all the QQUF nodes and I no longer have any problems.

    No_Cantaloupe_735599Nov 10, 2025· 3 reactions
    CivitAI

    wow, v.2 is awesome, take my buzz

    StawyNov 10, 2025
    CivitAI

    v2 is too good to be true.

    chrisc78Nov 12, 2025· 1 reaction
    CivitAI

    With the new V2 pose model Is there a way to control the size of the girls breasts? The lora is creating much bigger breasts then I would like it to. I am using the lora on Anime photos.

    The_Cook
    Author
    Nov 12, 2025

    V2 will have a good estimation of what the breast size should be, since it judges that from the i2v input, but you can always prompt for specific sized breasts. I think Wan 2.2 is usually smart enough to understand

    StawyNov 12, 2025

    @The_Cook v2 has a very good estimate of size and position, but not always. Sometimes it makes breasts larger than they should be (or smaller) and does not simulate gravity on breasts constricted by clothing. If there is a v3 in the future, it would be good to have more customisation with the prompt (breast size and shape, areola size and colour, etc.) and to fix the minor issues I mentioned. In any case, v2 is one of the best Loras for wan 2.2. Congratulations!

    GlowingGuardianGirlNov 12, 2025· 1 reaction

    @Stawy Or you leave OP alone and use the Wan breasts size slider 😂

    chrisc78Nov 14, 2025· 1 reaction

    Thanks, your suggestion to use a lora to control breast size works good.

    imjamesliquinNov 13, 2025
    CivitAI

    I am having an issue using the workflow you used. For some reason the FL_RIFE is missing. How can I fix this?

    The_Cook
    Author
    Nov 13, 2025· 1 reaction

    That node is part of Fill Nodes ( https://github.com/filliptm/ComfyUI_Fill-Nodes ). He re-wrote the RIFE frame interpolation so that it processes 10x faster and same quality than previous RIFE node implementations

    imjamesliquinNov 13, 2025

    @The_Cook Thank you!

    imjamesliquinNov 13, 2025

    @The_Cook I tried to install that node and it still shows up under the missing node filter. It still says that the FL_RIFE is missing. I tried different versions and its the same thing. Any idea?

    yofime9012866Nov 14, 2025

    where is the link for the workflow ?

    imjamesliquinNov 14, 2025

    @yofime9012866 Download one of the videos from the post and either drag and drop it or in comfyUI in the top left file>open.

    JabGalNov 13, 2025· 5 reactions
    CivitAI

    As "nudifier LORA" it works exceptionally well, one of the best I've tried. Congratulations!

    GardaXNov 15, 2025

    Nudify works without ANY Lora. No need Lora for that and it is not Lora making it. Just Wan itself.
    Just type "(Clothes from model disappear instantly:1.3)"

    altoiddealerNov 13, 2025
    CivitAI

    Thank you for this terrific LoRA!

    I think you should recommend what weights or range of weights to use these on for high and low. 1.0 and 1.0? etc.

    MMOFanNov 13, 2025
    CivitAI

    This is a nice LORA and really helps when you want the woman to turn around

    Rainart1989Nov 13, 2025· 2 reactions
    CivitAI

    Awesome lora!!! Thank you!!!

    For me, 4 out of 10 generations, the face changes

    I'm using wan 2.2 i2v rank64 lightx2v 4 step loras high 1030 and low 1022 both at 1.0 strength

    darkinlibiti2Nov 14, 2025
    CivitAI

    Best work! but have one problem is that faces change, quite dramatically, and they become completely similar. This is a huge problem, but it's the best lore that provides smoothness and quality. Unfortunately, it's also very important to preserve the face.

    The_Cook
    Author
    Nov 14, 2025

    I believe the faces changing is essentially an unsolvable problem with these types of loras. Your best bet is to try to keep the angle of the face the same, otherwise the model is guessing what the person looks like from a different angle. And that means the model is then creating the data from a guess. This problem is only solvable by introducing additional data such as an ipadapter for instance. It being a contextual i2v model also means that methods such as blurring faces or cropping does not solve this problem. The dataset already has lots of varied faces for it to work as a t2v model, as there are no repeated faces. You're going to see this complaint come up many times when you see these types of loras, and it will never be solved by the lora alone.

    darkinlibiti2Nov 15, 2025

    @The_Cook I’m using i2v, and the face angle there doesn’t change, but the face itself changes a lot when I use the LoRA. I’ve experimented a lot over a long time, and only playing with the LoRA weights helps to fix it, but not completely.

    Without this specific LoRA, the face is preserved almost 100% – the girl is posing, but not as beautifully as with your LoRA, you did an amazing job.

    It would be great if the LoRA could understand that it should only use the body and movement, but not the face, because this is a really big problem – specifically that when the LoRA is used, the face changes too much.

    Babbage78Nov 17, 2025

    @The_Cook I believe the way to get around this is to train a lora with a dataset that has had the faces of the women masked out.

    PPantsNov 14, 2025
    CivitAI

    I have issues where when the outfit changes it works for about a second and then the area that changed turns into a solid color block.

    a782776719716Nov 14, 2025

    me too!

    gomiPathNov 14, 2025

    The reason your outfit collapses one second after it changes is mostly because the Low setting is too strong.
    If you lower the Low value, it should become a lot more stable.

    The_Cook
    Author
    Nov 14, 2025

    make sure you're using the proper distillation loras along with this lora

    PPantsNov 14, 2025

    @gomiPath Im using str 1.0 for high and low@The_Cook with the linked distillation loras

    10280478Nov 15, 2025

    @PPants i2v seko 1 at str 1 for low, 1022 at str 0.8 or 1030 at str 0.5 for high. Euler or UniPC, shift 8-10. Use Painter I2V and 1030 together if you get opposite colored halo in place of clothes.

    linkzero7535999Nov 17, 2025
    CivitAI

    Greatness needs no words! The use of this Lora is beyond imagination! 

    aiartX_workNov 17, 2025· 3 reactions
    CivitAI

    Is it possible to strip clothing instead of simply disappearing.

    panzerknacker20111411Nov 18, 2025· 5 reactions

    yes. simple describe it in the prompt. like "She unzips her pants and slides them down her legs", or "she unknots her bikini-bottom which falls to the ground". it follows those kind of prompts quite good.

    fakelover2077Nov 21, 2025· 2 reactions
    CivitAI

    i love you man its so good pls keep making these and i hope you doing great in life! many thx <3

    FeathersMcGraw11Nov 29, 2025· 2 reactions
    CivitAI

    had great results with this, thanks!

    (for those experiencing face changed just prompt something like "their face stays the same, their expression is static" this will make them look a little doll-like but at least they will still look like them

    aiexplorer33Nov 29, 2025· 6 reactions
    CivitAI

    one of the best loras ever made. it captures the original body like no other lora i've seen. If you compare the first frame with clothes vs the second frame without clothes it's almost unnoticeable it's ai. it's mind blowing.

    smegmaripDec 30, 2025

    what he said...

    droeganDec 8, 2025
    CivitAI

    Would you (or someone else, really) be so kind as to point me towards the beta57 scheduler for comfyui?

    I can't seem to be able to locate it anywhere.

    LiliumComplexDec 8, 2025

    Get the "RES4LYFE" plugin for comfy, contains a variety of additional samplers and schedulers, among other things.

    jffddddddd5Dec 8, 2025· 3 reactions
    CivitAI

    this lora is even good for giving characters elegant movements, naked or not. very recommended

    ElectricalPotato61Dec 16, 2025· 1 reaction
    CivitAI

    This will be definitely you masterpiece! Best Lora ever! Thanks a lot!

    cai567890Dec 21, 2025· 8 reactions
    CivitAI

    Could you please not shave off pubic hair

    qqcoinslotDec 27, 2025· 9 reactions

    Just fap to hairy dudes instead, since pubic hair is not sex-specific. It literally conceals the goods.

    BioS_SlonFeb 6, 2026

    You can use Lora - wan21-wan22-pubic-hair. Adding hair is easier than removing it.

    axymeus991Dec 27, 2025
    CivitAI

    Would you share your training parameters? I'd love to be able to train hard cut transitions like that but can't get them to stick.

    Gekko78Jan 2, 2026
    CivitAI

    Just amazing. This is the missing piece i needed for my workflow. Much appreciated!

    smegmaripJan 29, 2026· 7 reactions
    CivitAI

    The lora was trained on a biased dataset of big boobs. 😂😂😂 Otherwise perfect.

    MilitAIMar 1, 2026· 1 reaction

    flat is justice!

    BioS_SlonFeb 1, 2026
    CivitAI

    The best lore. But I noticed there are problems when creating models other than the ideal model, like curvy or overweight ones. Is there a solution that works well with this lore?

    dread766Feb 7, 2026· 2 reactions
    CivitAI

    Why does this site not like its users...

    Search full name of this model, top result. Add in search filter for any of the wans this has, 57 results besides this one. Search for nude or posing with wan filter, one result....and it's not this lora.

    altaefrmpad990Mar 23, 2026

    SAME here - ALL the time!!!!!! Ridiculous.

    etharioMar 24, 2026
    CivitAI

    I'm pretty new at this but my program is saying the low noise 2.1 lora doesn't work with 2.2 what am I doing wrong ?

    LORA
    Wan Video 2.2 I2V-A14B

    Details

    Downloads
    15,827
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/9/2025
    Updated
    5/15/2026
    Deleted
    -

    Files

    W22_NSFW_Posing_Nude_i2v_LN_v2.safetensors

    Mirrors

    HuggingFace (61 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.