JSONify-Flux/README.md

---
license: apache-2.0
language:
- en
- zh
base_model:
- Qwen/Qwen2-VL-2B-Instruct
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- caption
- text-generation-inference
- flux
---
![8.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/3L4AaG9QB2A6fnoIQavt_.png)

# **JSONify-Flux**

The **JSONify-Flux** model is a fine-tuned version of Qwen2-VL, specifically tailored for **Flux-generated image analysis**, **caption extraction**, and **structured JSON formatting**. This model is optimized for tasks involving **image-to-text conversion**, **Optical Character Recognition (OCR)**, and **context-aware structured data extraction**.

#### Key Enhancements:

* **Advanced Image Understanding**: JSONify-Flux has been trained using **30 million trainable parameters** on **Flux-generated images and their captions**, ensuring precise image comprehension.

* **Optimized for JSON Output**: The model is designed to output structured JSON data, making it suitable for integration with databases, APIs, and automation pipelines.

* **Enhanced OCR Capabilities**: JSONify-Flux excels in recognizing and extracting text from images with a high degree of accuracy.

* **Multimodal Processing**: Supports both image and text inputs while generating structured JSON-formatted outputs.

* **Multilingual Support**: Trained to recognize text inside images in multiple languages, including English, Chinese, European languages, Japanese, Korean, Arabic, and more.

### How to Use

```python
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

# Load the model with optimized parameters
model = Qwen2VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/JSONify-Flux", torch_dtype="auto", device_map="auto"
)

# Recommended acceleration for performance optimization
# model = Qwen2VLForConditionalGeneration.from_pretrained(
#     "prithivMLmods/JSONify-Flux",
#     torch_dtype=torch.bfloat16,
#     attn_implementation="flash_attention_2",
#     device_map="auto",
# )

# Default processor
processor = AutoProcessor.from_pretrained("prithivMLmods/JSONify-Flux")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://flux-generated.com/sample_image.jpeg",
            },
            {"type": "text", "text": "Extract structured information from this image in JSON format."},
        ],
    }
]

# Prepare for inference
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# Generate output
generated_ids = model.generate(**inputs, max_new_tokens=256)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```

### JSON Output Example:
```json
{
  "image_id": "sample_image.jpeg",
  "captions": [
    "A futuristic cityscape with neon lights.",
    "A digital artwork featuring an abstract environment."
  ],
  "recognized_text": "Welcome to Flux City!",
  "metadata": {
    "color_palette": ["#FF5733", "#33FF57", "#3357FF"],
    "detected_objects": ["building", "sign", "street light"]
  }
}
```

### **Key Features**

1. **Flux-Based Training Data**  
   - Trained using **Flux-generated images** and captions to ensure high-quality structured output.  

2. **Optical Character Recognition (OCR)**  
   - Extracts and processes textual content within images.

3. **Structured JSON Output**  
   - Outputs information in **JSON format** for easy integration with various applications.

4. **Conversational Capabilities**  
   - Handles **multi-turn interactions** with structured responses.

5. **Image & Text Processing**  
   - Inputs can include **images, text, or both**, with JSON-formatted results.

6. **Secure and Optimized Model Weights**  
   - Uses **Safetensors** for enhanced security and efficient model loading.