I am reaching out to request technical clarification regarding model availability and endpoint integration within the Gemini Canvas runtime environment, (GEMINI WEB CANVAS, not AISTUDIO).
I am currently developing a web application (Single Page Application) running directly inside the Canvas UI. I have identified an inconsistency in how the Sandbox maps native endpoints across the Gemini 3.1 image generation model family:
1. `Gemini 3.1 Flash Image`: Executes seamlessly by consuming the active user’s authenticated web session context without requiring explicit authorization headers or credential injection.
2. `Gemini 3.1 Flash-Lite Image` & `Gemini 3.1 Pro Image`: Execution fails when invoked through the local Canvas client unless requests are explicitly signed with an API key managed via Google AI Studio or Vertex AI (`x-goog-api-key`).
Specific technical questions:
Runtime Architecture: Is there a technical constraint preventing the Canvas internal bridge (*mock/context SDK*) from implicitly exposing the active user’s quota for the Flash-Lite variant, given its lower computational overhead and reduced latency profile compared to the standard Flash model?
Integration Roadmap: Are there plans to unify access across the entire model family (Lite, Flash, and Pro) under the native Gemini session token, eliminating the need for end users to supply developer credentials for applications running inside the Canvas ecosystem?
Subscription-based Quota Allocation: For users with an active paid tier (e.g., Gemini Advanced / Google One AI Premium), is there any technical barrier preventing Canvas from automatically granting native access to the Pro and Lite endpoints under the user’s paid quota, avoiding the requirement of manual API key configuration?
To make it clear that you understand what I’m talking about, I leave this app that does a test precisely of what I’m talking about.
I can understand that you don’t want to let us use the Pro model without an API inside the Gemini canvas itself, but that you don’t let us use the Lite model? I really don’t understand that.
By using them within Gemini and the canvas, in any case, google deviate us from the credit of our subscription.
So I don’t see the problem of not being able to also enable the Lite model.
Thank you for your time and for clarifying these technical specifications and execution boundaries.