CivArchive
    Summer Cummings - v0.4
    NSFW
    Preview 1
    Preview 2
    Preview 3
    Preview 4
    Preview 5
    Preview 6

    Hi, this is a Kohya trained LoRA model for 90s pornstar actress Summer Cummings

    v0.5 (SD1.5):

    • dataset with 72 images (better enhanced with Topaz Photo AI
      without going overboard with the upscale, then tagged with wd-v1-4-swinv2-tagger-v2 then captioning reviewed manually)

    • changed some parameters in kohya: network dim always at 128 but network alpha lowered at ??, lr_scheduler changed from "constant" to "cosine with restarts", clip skip from 2 to 1

    At this point I'm quite satisfied, but I'm already thinking about how to improve the model, obviously I'll apply everything I've learned to give life to other 90s pornstars!

    Description

    Not too happy with v0.3 I decided to make a couple of changes in the training settings:

    • in the captioning I avoided the word "dark" (whereas for example when I described the hair color as "dark brown hair"), from the img generations with ReV Animated I noticed that it had the tendency to create environments, precisely, dark;

    • in addition I decreased the batch size from 2 to 1, and given the results I think a little slowness was needed in learning

    I'm sorry that this can be seen as a work in progress, but actually the intent is to compare myself with other creators to better understand how to get good results.
    In the next version I'll use regularization images, never used by me until now to create LoRA models of a person. When I've used them in the past with DreamBooth to create a full checkpoint the results have been better than those obtained so far.

    FAQ

    LORA
    SD 1.5

    Details

    Downloads
    1,337
    Platform
    SeaArt
    Platform Status
    Available
    Created
    4/2/2024
    Updated
    9/25/2025
    Deleted
    -
    Trigger Words:
    smm

    Files

    Available On (2 platforms)

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