public/ai/provider/openai/UPGRADING.md
A new aiprovider_openai\aimodel\openai_image_base interface has been added. Image generation model classes must now implement this interface to declare their response_format, output_format, size, and quality mappings. Existing custom model classes that handle image generation should implement this interface to ensure correct API parameters are sent.
For more information see MDL-85352
A new gptimage1 model class has been added to support gpt-image-1.5.
This model uses output_format=png instead of response_format, and maps Moodle quality values to the values expected by the API: 'standard' maps to 'medium' and 'hd' maps to 'high'.
For more information see MDL-85352
The dalle3 model class now implements openai_image_base and switches from returning a URL to returning response_format=b64_json.
The image is now decoded directly from the API response instead of being downloaded via a second HTTP request. Size and quality logic has been moved into the model class.
For more information see MDL-85352