IMPORTANT: READ BELOW
if you want, can add prompt "1990s \(style\)" (small change)
Trigger word(s): ctianchid0ri, blue hair,
Clothing: blue hair, white blazer, red bowtie, ribbon, long sleeves, blue skirt,
commission: https://ko-fi.com/ctian 😼
Example
Positive prompt ✅:
score_9, score_8_up, score_7_up, source_anime, masterpiece, 1girl, ctianchid0ri, blue hair, white blazer, red bowtie, ribbon, long sleeves, blue skirt, looking at viewer, outdoors, city, depth of field, sunny, black kneehighs, Negative prompt ❌:
Recommended resolution, aspect ratio for sdxl, pony 📷
768 x 1344: Vertical (9:16)
832 x 1216: Portrait (4:5) **
1024 x 1024: Square (1:1)
1182 x 886: Photo (4:3)
1254 x 836: Landscape (3:2)
1365 x 768: Widescreen (16:9)
Work flow, procedure 🔧
1. Collect dataset (booru, anime screencaps, search)
2. Clean up dataset using upscalers, inpainting and photoshop
3. Tag dataset using booru github
4. Train LoRa in kohya ss locally github
5. Test epochs using xyz grid in controlnet to ensure LoRa is not overfitted/underfitted.
6. Submit model to CivitAI profile
do not ask me training parameters, check lora metadata for training parameter