Trained on NAI. V1.5:Best weight between : 0.75-0.95.<pre><code>Prompt activator: ikezawa hanako
Major prompt words: burn mark, burn scar, scar, hair over one eye</code></pre>Outfit prompts:<pre><code>school uniform: school uniform, black bow, shirt, long sleeves, green skirt</code></pre><pre><code>casual outfit: casual outfit, berret, black bowtie, jacket, black pants</code></pre><pre><code>nightgown: nightgown</code></pre>I added images to the dataset and reorganized the prompt words. I think that, perhaps, textual inversions like 'bad hand' reduce burns on the hand. The burns look more similar, even though they still vary somewhat. You can increase their prominence by adjusting the LoRA weight or by using parentheses, for example, '(burn scar:1.1)'. The outfits are a bit more accurate. In some complex prompts, you may need to add the clothes color. I initially included the color in the prompt to avoid any issues, but you can try removing it to make the prompt lighter if you prefer.V1.0 Prompt activator:  hanakoikezawa Major prompt words: hair over one eye, burn Outfit prompts:<ul><li>school uniform, (green skirt, black tights, black bow, long sleeves)*</li><li>main outfit, (black hat, blue jacket, black pants)*</li><li>nightgown</li></ul>*If the main trigger word does not work properlyBest weight between : 0.7-0.95.It's hard to have her face burned without using 'hair over one eye.' You can increase the trigger word 'burn' or ask for the face to be specifically burned, but then the entire face will also be affected, not just the right side alone. I am showing two examples in my preview images.If you have any issues with this LoRA or if you have any suggestions, don't hesitate to communicate them.

Hanako Ikezawa (Katawa Shoujo)

I added images to the dataset and reorganized the prompt words. I believe that textual inversions like 'bad hand' reduce burns on the hand. The burns look more similar, even though they still vary somewhat. You can increase their prominence by adjusting the LoRA weight or by using parentheses, for example, '(burn scar: 1.1)'. The outfits are a bit more accurate. In some complex prompts, you may need to add the clothes color. I initially included the color in the prompt to avoid any issues, but you can try removing it to make the prompt lighter if you prefer.

HanakoIkezawa2.safetensors

v1.5

v1.0

V1.5:Best weight between : 0.75-0.95Prompt activator: ikezawa hanako Major prompt words: burn mark, burn scar, scar, hair over one eyeOutfit prompts:school uniform: school uniform, black bow, shirt, long sleeves, green skirtcasual outfit: casual outfit, berret, black bowtie, jacket, black pantsnightgown: nightgownI added images to the dataset and reorganized the prompt words. I think that, perhaps, textual inversions like 'bad hand' reduce burns on the hand. The burns look more similar, even though they still vary somewhat. You can increase their prominence by adjusting the LoRA weight or by using parentheses, for example, '(burn scar:1.1)'. The outfits are a bit more accurate. In some complex prompts, you may need to add the clothes color. I initially included the color in the prompt to avoid any issues, but you can try removing it to make the prompt lighter if you prefer.

I added images to the dataset and reorganized the prompt words. I think that, perhaps, textual inversions like 'bad hand' reduce burns on the hand. The burns look more similar, even though they still vary somewhat. You can increase their prominence by adjusting the LoRA weight or by using parentheses, for example, '(burn scar:1.1)'. The outfits are a bit more accurate.

In some complex prompts, you may need to add the clothes color. I initially included the color in the prompt to avoid any issues, but you can try removing it to make the prompt lighter if you prefer.

Files with SHA256: ac50737744da2a5aa0d628251c155100dd59291e8d4a8899f4b3e8483e884efc