🐱 LoRA Fine-tuned Kontext Model for Cat-Face Food Transformation
Repository: https://huggingface.co/svjack/Kontext_Cat_Face_Lora
This repository provides a LoRA (Low-Rank Adaptation) fine-tuned variant of the Kontext image consistency model, specialized in adding adorable cat faces to food items while preserving original textures, lighting, and structural details.
✨ Key Features
Texture & Detail Preservation
- Retains original food textures (e.g., bun crispiness, broth glossiness) while seamlessly integrating cat facial features
- Maintains lighting consistency and material properties (e.g., glass translucency, sauce shine)
Low-Rank Adaptation Efficiency
- Optimized for fast inference with minimal GPU memory usage
- Compatible with consumer-grade hardware (e.g., NVIDIA 16GB+ GPUs)
Prompt Flexibility
- Simple prompts like
make food in cat style, add ears, glasses, beard, mouth, tail.trigger precise edits - Supports multi-object food compositions (e.g., stacked burgers, liquid dishes)
- Simple prompts like
🍜 Transformation Examples
Example 1.1: Glass Peach
| Original Food | Transformed Output |
|---|---|
 |
 |
| Preserved: Peach translucency, stem texture, glass refraction |
Example 1.2: Glass Pear
| Original Food | Transformed Output |
|---|---|
 |
 |
| Preserved: Pear translucency, stem texture, glass refraction |
Example 2: Hamburger
| Original Food | Transformed Output |
|---|---|
 |
 |
| Preserved: Sesame seed distribution, cheese melt, lettuce folds |
Example 3: Noodles
| Original Food | Transformed Output |
|---|---|
 |
 |
| Preserved: Broth shimmer, noodle strand separation, steam effects |
Example 4: Adeptus' Temptation (Genshin Impact)
| Original Food | Transformed Output |
|---|---|
 |
 |
| Preserved: Glaze glossiness, ingredient layering, golden glow |
⚙️ Technical Details
- Base Model: Kontext (Flux.1 Kontext-dev, 12B rectified flow transformer)