OpenAI image generation support

[amitsnow]

Summary

Adds complete text-to-image generation and image editing support for DALL-E-2, DALL-E-3, and GPT-Image-1 models with unified output_type: image interface, auto-detection of operation type, multi-image support, and automatic file saving.

Features Implemented

Text-to-Image Generation:

• DALL-E-2: Batch generation (up to 10 images), sizes 256x256-1024x1024
• DALL-E-3/GPT-Image-1: HD quality, vivid/natural styles, sizes up to 1792x1024

Image Editing:

• DALL-E-2: Single image editing
• GPT-Image-1: Multi-image editing (1-16 images)

Output Processing:

• Returns base64 data URLs (single string or JSON array for n>1)
• Auto-saves to multimodal_output/image/ directory
• Replaces data URLs with file paths in output

Performance Impact

• Single-pass processing, lazy directory creation
• Negligible overhead (I/O bound by API calls)

How to Test

Basic Testing:

1. Text-to-image: Set output_type: image, provide text prompts
2. Batch generation: Set n: 3 in parameters (DALL-E-2)
3. Image editing: Provide images in prompt with text instructions
4. Multi-image editing: Use GPT-Image-1 with 2-16 images

Unit Tests:

pytest tests/core/models/test_custom_openai.py -k "image" -v  # 16 tests
pytest tests/utils/test_multimodal_processor.py -v             # 10 tests

Configuration Example:

gpt_image_1:
  model: gpt-image-1
  output_type: image 
  model_type: azure_openai  
  api_version: 2025-04-01-preview
  parameters:
    size: "1024x1024" 
    quality: "high" 

Example Output

Single image (n=1)
{"id": "1", "image": "multimodal_output/image/1_image_0.png"}

Multiple images (n=3)
{"id": "1", "images": ["...image/1_images_0.png", "...1_images_1.png", "...1_images_2.png"]}

Checklist

Checklist

• Lint fixes and unit testing done
• End to end task testing
• Documentation updated

[psriramsnc]

LGTM 🚀