metadata
license: apache-2.0
base_model: Qwen/Qwen-Image-Edit-2509
tags:
- image-editing
- vision
- multimodal
- zen
- zoo-gym
- hanzo-ai
- text-to-image
- image-to-image
language:
- en
- zh
pipeline_tag: image-to-image
library_name: transformers
Zen-Image-Edit 🎨
Part of the Zen AI Model Family | Based on Qwen-Image-Edit-2509
✨ Model Highlights
Advanced 7B image editing model with natural language instructions:
- Object Manipulation: Add, remove, move objects seamlessly
- Style Transfer: Apply artistic styles and filters
- Background Editing: Replace or modify backgrounds
- Face Editing: Adjust expressions and features
- Resolution: Up to 1024x1024
- Speed: 3-5 images/second on A100
📊 Performance
| Benchmark | Score |
|---|---|
| EditBench | 87.3% |
| MagicBrush | 82.1% |
| InstructPix2Pix | 89.5% |
| ImageNet-E | 91.2% |
💻 Quick Start
from transformers import AutoModelForImageEditing, AutoProcessor
from PIL import Image
model = AutoModelForImageEditing.from_pretrained("zenlm/zen-image-edit")
processor = AutoProcessor.from_pretrained("zenlm/zen-image-edit")
image = Image.open("input.jpg")
instruction = "Remove the car and add trees"
inputs = processor(images=image, text=instruction, return_tensors="pt")
edited_image = model.generate(**inputs)
edited_image.save("output.jpg")
🎨 Editing Capabilities
- Object Removal: Clean inpainting with context awareness
- Object Addition: Natural placement with proper lighting
- Style Transfer: Artistic transformations
- Color Grading: Professional color adjustments
- Background Swap: Seamless background replacement
- Face Editing: Expression and feature modification
- Weather Effects: Add rain, snow, fog
- Time of Day: Convert day to night scenes
📦 Available Formats
| Format | Size | Use Case |
|---|---|---|
| SafeTensors | 14GB | Full precision |
| GGUF Q8 | 7GB | High quality |
| GGUF Q4 | 3.5GB | Mobile/edge |
| MLX 8-bit | 7GB | Apple Silicon |
| MLX 4-bit | 3.5GB | iOS devices |
Built by Hanzo AI × Zoo Labs Foundation