Diffusion-KanjiRadComp

Diffusion-KanjiRadComp is a fine-tuned Stable Diffusion model trained on 200 curated Kanji samples.
The dataset is balanced at a 1:1 ratio between radicals (basic structural components of kanji) and their applications (full kanji that incorporate those radicals).

The model was fine-tuned using LoRA (Low-Rank Adaptation) on top of CompVis/stable-diffusion-v1-4 to improve generation of novel kanji from English text prompts.

---

Training Details

• Dataset: 200 samples (100 radicals + 100 compound kanji)
• Image resolution: 128×128
• Captions: English meaning(s) only
• Base model: CompVis/stable-diffusion-v1-4
• Fine-tuning method: LoRA

---

Sample Generation

Example prompt:
prompt = "fire"

Prompt

Screenshot