Imagen 4 API models hit their shutdown date — migrate to Gemini
Three imagen-4.0 endpoints reach Google's earliest shutdown day; image work shifts to Gemini's 'Nano Banana' models. Plus Copilot CLI shared sessions.
Copy markdown
The three IDs that stop serving
imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 reach August 17 — the earliest date Google will pull them from the Gemini API. Hardcoded calls to those endpoints will start failing, so audit your image pipeline now.
Where the work moves: Gemini's native image models
Google routes you to in-Gemini image generation — the 'Nano Banana' line, such as gemini-3.1-flash-image. There's no separate migration fee; you pay standard Gemini API image pricing and quota. It's a weekend swap: point at the new model and re-run your prompt suite.
Re-test before you route production traffic
Don't assume prompt parity. Google flags differences in aspect ratios, safety handling, latency, quota, and cost between Imagen 4 and its Gemini image models — validate visual quality and prompt adherence on your own cases before cutting over.
Elsewhere: Copilot CLI gets shared host sessions
GitHub's Copilot CLI (v1.0.80) added Agent Host Protocol sharing — multiple terminals can attach to one session and watch its turns stream live. Useful for pairing on, or babysitting, a long-running agent from a second window.
Also new: agent apps land inside your PRs
GitHub added agent-app integrations — Amplitude, Endor Labs, LaunchDarkly, and PagerDuty — surfacing analytics, dependency scans, feature flags, and deploy-risk checks directly in a pull request, with no context switch.