LoRA model for recreating the features of I-47 (a.k.a. Yona, 伊47, ヨナ) from the browser game and media franchise Kantai Collection.
I trained this because I was having much difficulty in recreating Yona's accessories and outfit, even when using models trained on sets with her in it (she has too many alternate outfits). ヨナちゃんかわいい
I recommend using model weights between 0.6-0.8. It's also possible to go higher for "better" feature fidelity and img2img afterwards to fix the quality.
Prompting
Activation token "yonachan" can be used to strengthen her features.
The training set consisted entirely of images in her default outfit, so although you can suppress her usual outfit in the negative prompt (see example image metadata), any clothes on her will tend to behave like her usual outfit (skintight, collars, swimsuits under clothing, etc.). Try to lower the model weight as much as possible if you're prompting for alternate outfits.
If there's enough interest I may make a v3 with more versatility.
Description
Example images
Main images are generated in txt2img and then upscaled by 2x using Ultimate SD Upscaler.
All images were generated using AOM2-nutmegmixGav2.
Negative prompts utilizes bad-artist and bad-artist-anime embeddings from https://huggingface.co/NiXXerHATTER59/bad-artist
Dataset Info
59 manually pruned images of Yona in her default outfit from Danbooru. Images were cropped and/or cleaned to remove signatures and undesired text and (most) other characters.
Tagging was done via wd14-swinv2-v2 with threshold 0.15, with "yonachan" added in front of each list.
Training parameters
Trained on AOM2-nutmegmixGav2 with 20 repeats per image.
"net_dim": 128,
"alpha": 128.0,
"scheduler": "cosine_with_restarts",
"cosine_restarts": 1,
"warmup_lr_ratio": 0,
"learning_rate": 0.00032,
"text_encoder_lr": 6e-05,
"unet_lr": 0.00032,
"shuffle_captions": true,
"keep_tokens": 1,
"train_resolution": 768,
"lora_model_for_resume": null,
"unet_only": false,
"text_only": false,
"vae": null,
FAQ
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.