Mink is the main character of Dragon Half.The main trigger is <code>mink dragon_half</code>, with either <code>dragon horns</code> or <code>multiple horns</code> to specify a variant.Default anime appearance: <code>mink dragon_half, dragon horns, bikini armor, detached sleeves, vambraces, dragon wings, dragon tail, knee boots, dragon charm</code>. Weight 0.9 is recommended.<h3 id="activation-tags">Activation tags</h3>1) <code>mink dragon_half</code> is the main trigger. This won't add any horns, just Mink with her spiky hair and red eyes. Add it at the start of your prompt. 2) <code>dragon horns</code> or <code>multiple horns</code>. Those two tags were not compatible with each other during training and testing so I specifically separated them. The <code>dragon horns</code> tag adds a standard pair of horns, like in the anime. While <code>multiple horns</code> is used to add the 3 pairs of horns from her post-transformation manga appearance. 3) <code>dragon charm, bikini armor, detached sleeves, vambraces, knee boots</code> creates the outfit Mink is wearing in the anime. I spent a lot of time training each keyword separately, so you should be able to mix and match them, although you might need to add them to the negatives to stop them from bleeding into each other. 4) <code>dragon wings</code> and <code>dragon tail</code> are self explanatory.Bonus tags: a) <code>anime screenshot</code> makes the generation look like screenshots from the official anime. b) <code>official art</code> can be used to shift the art style toward the official illustrations.<h3 id="misc.">Misc.</h3>Preparing the dataset and training took a lot of time. I went over every image and manually edited them to unify the appearance, and create multiple variants for higher flexibility. All to hopefully make this LoRA as versatile as possible.Trained on about 250 images, separated into subsets for no horns, dragon horns, and multiple horns. 13 epochs (2200 steps total), 0.00006 learning rate, 8 DIM / 4 Alpha, Came, cosine with restarts 3.Feedback is welcome.

Mink (Dragon Half)

mink_illustrious-4as.safetensors

v1.0

Mink is the main character of Dragon Half.The main trigger is <code>mink dragon_half</code>, with either <code>dragon horns</code> or <code>multiple horns</code> to specify a variant.Default anime appearance: <code>mink dragon_half, dragon horns, bikini armor, detached sleeves, vambraces, dragon wings, dragon tail, knee boots, dragon charm</code>. Weight 0.9 is recommended.<h3 id="activation-tags">Activation tags</h3>1) <code>mink dragon_half</code> is the main trigger. This won't add any horns, just Mink with her spiky hair and red eyes. Add it at the start of your prompt. 2) <code>dragon horns</code> or <code>multiple horns</code>. Those two tags were not compatible with each other during training and testing so I specifically separated them. The <code>dragon horns</code> tag adds a standard pair of horns, like in the anime. While <code>multiple horns</code> is used to add the 3 pairs of horns from her post-transformation manga appearance. 3) <code>dragon charm, bikini armor, detached sleeves, vambraces, knee boots</code> creates the outfit Mink is wearing in the anime. I spent a lot of time training each keyword separately, so you should be able to mix and match them, although you might need to add them to the negatives to stop them from bleeding into each other. 4) <code>dragon wings</code> and <code>dragon tail</code> are self explanatory.Bonus tags: a) <code>anime screenshot</code> makes the generation look like screenshots from the official anime. Warning: Not compatible with Pony's quality tags - you will have to decrease their strength! b) <code>official art</code> can be used to shift the art style toward the official illustrations. Also not compatible with Pony's quality tags - you will have to use something like this: <code>(score_9, score_8_up, score_7_up, score_6_up:0.7)</code><h3 id="misc.">Misc.</h3>Preparing the dataset and training took a lot of time. I went over every image and manually edited them to unify the appearance, and create multiple variants for higher flexibility. All to hopefully make this LoRA as versatile as possible.Trained on about 250 images, separated into subsets for no horns, dragon horns, and multiple horns. 25 epochs (4000 steps total), 0.0003 learning rate, 8 DIM / 4 Alpha, AdamW8bit, cosine with restarts 3, 0.00/0.00/0.01 dropouts.Feedback is welcome.

Files with SHA256: 7089c916ea9a959590e5ac17f6e0a3d09675156fbfdd1084a7eb53151e36f05d