Image generation usually arrives as a separate integration with its own keys and no audit. This release makes it a model type of the platform, configured, selected and recorded the same way as the chat and embedding models already are.
A model type like the others
- Configured centrally. Image models are declared in the platform's AI provider configuration beside chat and embedding models, and resolved from it at call time.
- Two providers. OpenAI image models and Google Imagen on Vertex AI are supported.
- Audited without the payload. Every generation is recorded in the AI audit with its model, caller and outcome; the image data itself is kept out of the log.
Where it is used
- From server-side scripts. Scripts can generate an image through the platform's AI namespace and store the result directly in the database.
- The caller's default. Features that generate images use the calling user's default image model, so the choice of provider stays an administrative decision.
No image model is enabled until an administrator configures one; existing chat and embedding configurations are unaffected.