CivArchive
    Nikki Charm 80s Porn Star Wan 2.1/Hunyuan - Wan v1.0

    For some reason I'm not seeing Wan character loras. Is there a reason? Tell me if there is. I think I may have been first, but the flood gates appeared to have opened. Great to see mor Wan character loras! Wan v1.0 is trained on 14b at 480p. Samples are upscaled 2x.

    Hunyan V1 was trained at 512x512, V2 used the same dataset, but trained at 768x768. Samples for Hunyuan V2 are 448x720, and not upscaled. Same prompts.

    Shannon Louise Eaves born 1966, active 1984 to 1990 under the name Nikki Charm. According to IMDb she was 4ft 11 inches, and Blonde, adorable, enthusiastic and cute as a button, she always had a big grin.

    Lora trained using ttplanet's Windows GUI for Kohya's musubi-tuner. Trained mostly from screencaps of very low resolution videos from the 80s. 72 images, 120 epochs

    Samples produced with hunyuan_video_FastVideo_720_fp8_e4m3fn, 400x640, 8 steps, for 121 frames. 4x vti multiplier using Rife Tensort and rife49. Upscaled 2x with Topaz Video.

    Prompts used:

    nikkic, closeup video of a young woman with platinum blonde hair in a ponytail, on the sidelines of a football game. She is wearing a red and white cheerleader uniform.

    nikkic, closeup video of a young woman with platinum blonde hair, standing on the pitcher's mound, holding a microphone and singing the national anthem at a baseball game.

    nikkic, closeup video of a young woman with fluffy platinum blonde hair, walking slowly towards the viewer on a path through a park on a summer day,looking from side to side. she is wearing business casual

    nikkic, closeup video of a young woman with fluffy platinum blonde hair, having lunch in a coffee shop. she is wearing business casual

    nikkic, closeup video of a young woman with platinum blonde hair in a ponytail, having lunch in a coffee shop. she is wearing business casual

    Description

    FAQ

    Comments (14)

    kmdcompMar 17, 2025· 3 reactions
    CivitAI

    I've seen a few people mentioned that getting a good likeness is harder in WAN, to the point of wondering if the model has actually been gimped in some way for character training. I plan to put that to the test tonight, because yours looks exactly how I remember her when I was getting to that age in the late 80's and my buddy had a tape...

    It's absolutely insane how good these models are at just "figuring it out", when it comes to low rez images or even low rez and low quality images.

    The conventual wisdom from SD1.5 and maybe SDXL (don't know, never trained it), that you must use HQ, high rez images, is long dead, as far as I can tell. I've been able to get very good LoRAs from from some truly terrible datasets with flux and hunyuan. Like made for TV movies in standard def from the 80's, ripped from a bad VHS tape, levels of bad.

    jpXerxes
    Author
    Mar 17, 2025· 1 reaction

    I agree that it is amazing what these models can do. There isn't an image that started out at more than 640x480 in original form, and some as low as 320x240. Generally upscaled with Topaz with face restoration, but that's it. Note that I took 12000 steps to get it to where I liked it.

    jm112368767Mar 18, 2025

    I've made several character LoRAs for my own purposes and I have to say it does seem harder to get WAN LoRAs to really look a lot like the person, as opposed to Hunyuan. Could certainly be my training process is lacking, but I've had great success with Hunyuan but very little with WAN.

    ltcdrbroccoliMar 18, 2025· 1 reaction
    CivitAI

    Excellent, I'm familiar with her, uh, work... Will have to try this out in WAN.

    jpXerxes
    Author
    Mar 18, 2025· 1 reaction

    Tell us if it lives up to your memories!

    MrToonMar 18, 2025
    CivitAI

    Nice work, pretty much the first Lora based on a real person i've seen on WAN... I was kinda worried WAN had a huge shortcoming. Is this based on https://github.com/tdrussell/diffusion-pipe or did you use something else?

    jpXerxes
    Author
    Mar 18, 2025· 1 reaction

    I'm using Kohya's musubi-tuner. https://github.com/kohya-ss/musubi-tuner I had been using ttplanet's GUI to train Hunyuan, but I found the Wan version to be too buggy, so I started using hand-formed command lines. A couple of days ago I wrote my own GUI to make it easier. Before anyone asks, it is an executable integated with my own file structure, so I won't be distributing it

    WazzieMar 18, 2025
    CivitAI

    Nice work, how did you divide the video and the image datasets? I am currently working on some character loras, the results are okay, but not as good as yours.

    jpXerxes
    Author
    Mar 19, 2025· 1 reaction

    strictly images, although many were video caps

    herosavesthedayMar 18, 2025
    CivitAI

    I'd imagine that you aren't seeing many character LoRAs for Wan because the i2v is so good.

    MrToonMar 18, 2025· 1 reaction

    now imagine i2v with a lora of that character

    2988Mar 19, 2025· 1 reaction

    @MrToon yeah true, because once the character in img2vid starts turning around etc, or you want to continue the video from last frame, things get rea; messy, a character lora can solve that

    vbobMar 18, 2025
    CivitAI

    Colleen Brennan when?

    jpXerxes
    Author
    Mar 19, 2025

    They had similar looks, but Colleen was 17 years older. Are you asking for a lora? Not sure I have any source material

    LORA
    Wan Video

    Details

    Downloads
    292
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    3/17/2025
    Updated
    5/11/2026
    Deleted
    5/23/2025

    Available On (2 platforms)

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