CivArchive
    Gabrielle Union [SDXL] - v1.0
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    An impression of American Actress Gabrielle Union

    Born: October 29, 1972 - Omaha, Nebraska

    I've had the best luck with the weight around 1.0-1.3 and CFG between 2-5

    SD 1.5 model of Gabrielle Union:
    Gabrielle Union [SD 1.5] - v1.0 | Stable Diffusion LoRA | Civitai

    Description

    FAQ

    Comments (4)

    astdioJun 20, 2024
    CivitAI

    V1.0 doesn't really work for me. The trigger word does nothing, probably because it is also part of all the reg images.
    When used in combination with other tags (woman, looking at viewer, lips, long hair) it seems to come together, but the resemblance was still hit or miss for me.

    LimitationsUndone
    Author
    Jun 20, 2024

    Thanks for the feedback. I'm still going through trial and error with the SDXL models. I'm still more at ease with SD 1.5 models. I'll try out a reg folder other than the trigger word on future models to see if that works. Right now, I'm working with a 6GB card and training SDXL takes about 20 hours so not a quick turnaround.

    astdioJun 20, 2024

    @LimitationsUndone Combination of tags will work. So if you combine the trigger word with "woman, looking at viewer, long hair, dark hair" it often works. It seems the more tags you use from your training data the better.
    I don't know what you used for your reg images, but I usually don't tag them and I don't use the trigger word in them, that will only mess up things. For SDXL I normally don't bother with reg images anyway.

    LimitationsUndone
    Author
    Jun 22, 2024

    @astdio Reg images I use are different body images of the subject, in this case various body images of Gabrielle Union. I feel reg images of the model, at least when I do my SD 1.5 models, represents them better than the checkpoint's representation. I tag them because I was getting weird results by not doing it. My main training images are just headshots from various angles and expressions. For my preview images, I just use a few of the templates I've made changing the trigger word. But do make some unique images from the tagged words that, for me, seem to work. Here is my approach to making a model, only slight modifications like training with a SDXL checkpoint and going from 512x512 to 1024x1024:
    My Approach to Making a Celebrity LORA | Civitai

    LORA
    SDXL 1.0

    Details

    Downloads
    108
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    6/19/2024
    Updated
    5/8/2026
    Deleted
    5/23/2025
    Trigger Words:
    64bri3113uni0n

    Files

    Gabrielle_Union_PMv1_SDXL.safetensors