Type | |
Stats | 408 |
Reviews | (46) |
Published | Sep 16, 2024 |
Base Model | |
Training | Steps: 423 Epochs: 6 |
Usage Tips | Clip Skip: 1 |
Trigger Words | kimberlyhart young woman slender delicate face in late teens to early twenties light to medium skin tone light-brown hair |
Hash | AutoV2 C0149483C2 |
UPDATE (10/23):Fellow MMPR fan + seasoned Lora maker https://civitai.com/user/Peli86 contacted me with some advice on better dataset making, and since I wasn't fully satisfied with how close the Flux lora got likeness-wise, after a few attempts, I'm pretty happy with this one! Will update with more sample images and prompts to replicate source outfits accurately. For now, I do recommend keeping strength around 0.7-0.8 and specifying that she's more skinny/slender because generation always seems to automatically buff her up for some reason.
UPDATE (10/17): Wasn't satisfied with the initial XL model, then remembered I had a lot of additional and higher-resolution training images from the Flux that I could bring over. Also learned how to get better image quality with them in ComfyUI so check the workflow for those. The original XL is still available but not recommended.
Now with an XL equivalent! Note the different trigger words and I recommend opening both LORAs in NotePad to see a list of outfits.
It's nostalgia time. As played by Amy Jo Johnson, and trained using various upscaled screengrabs from Mighty Morphin and HD versions of both 90s theatrical films.
I'm still trying different settings and inputs to better recreate the outfits and hairstyles I used in training and will update this description and the sample photos with stuff so everyone can know the right text to depict them accurately. For now will definitely recommend including "buff" in the negative prompts because it really likes to depict her as taller and stockier than she actually is.