Trained entirely on screencaps, the below noted outfit is directly from the anime
Works with other lora
Recommended weight = 0.8 - 1
The Showcase images use weight 1
Base prompt - combine with below noted outfits or use other outfit prompts
TNMOgawa, brown hair, long hair, braid, single braid, hair over shoulder, brown eyes, blunt bangs, braided ponytail,
School Outfit
OgawaSchool, bow, white shirt, shirt, bowtie, school uniform, black bow, black bowtie, long sleeves, skirt, pantyhose, black pantyhose, underbust, collared shirt, pleated skirt, vest, yellow vest, dress shirt, buttons, hair ribbon, ribbon, high-waist skirt, brown skirt, miniskirt, black ribbon,
Any kind of feedback on my models is highly appreciated!
I love seeing the images made with my models, dont be shy, post some!
Commission a character here !!
Description
Comments (5)
Why does the Pony version look so much prettier than the Illustrious version?
@lentor I recently adjusted the steps/repeats calculation so there's less overlap with outfits having a ton of images and other outfits with only a few images. This has also (positively) impacted characters with a very small dataset. The change was implemented after the illustrious model was trained, so the model didn't get that "upgrade" yet.
I can re-train the illustrious model if you want to use it. Please let me know :)
@KurutAI That would be nice. I've noticed that there are so many Pony characters that look better, but that's not always the case. If you have free time for this, I will be happy to see the difference for clarity.
@lentor depends heavily on how the training is configured. If both have the same training configuration, Illustrious is generally better at recreating outfits, while pony is better at creating images that follow prompts. At least in my opinion ^^
I'll re-train it once my currently queued batches are done, shouldn't take more than a few days
@KurutAI Cool!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.