Google is shutting down all Imagen 4 API endpoints on August 17, 2026. This affects Imagen 4 Standard, Imagen 4 Ultra, and Imagen 4 Fast across both the Gemini Developer API and the Agent Platform (formerly Vertex AI). If your application generates images through any of these endpoints, you have 10 days to migrate.
The replacement is the Gemini image generation family, internally called “Nano Banana” in Firebase documentation, and publicly referred to as “Gemini 3.1 Flash Image” in Google’s developer docs.
This guide covers what is changing, what replaces it, and how to migrate your code before the shutdown date.
What is shutting down
Three Imagen 4 endpoints are being deprecated and will stop accepting requests on August 17, 2026:
- Imagen 4 Standard - The general-purpose image generation endpoint used by most developers
- Imagen 4 Ultra - The high-fidelity endpoint for production-quality imagery
- Imagen 4 Fast - The low-latency endpoint optimized for speed over quality
This shutdown applies across Google’s entire AI platform:
- Gemini Developer API (ai.google.dev) - Imagen 4 models will no longer be available
- Agent Platform / Vertex AI (cloud.google.com) - Same models, same shutdown date
After August 17, any API call to an Imagen 4 model ID will return an error. There is no grace period or soft deprecation window beyond the announced date.
What replaces Imagen 4
Google’s replacement is the Gemini image generation model family. The naming is inconsistent across Google’s documentation:
- Google’s AI developer docs reference “Gemini 3.1 Flash Image” as the migration target
- Firebase documentation refers to the replacement models as “Nano Banana models”
Both refer to the same underlying technology. The Gemini image models are multimodal generation models that handle text-to-image, image-to-image, and image editing tasks within the broader Gemini model architecture.
This is a significant architectural shift. Imagen was a standalone image generation system. Gemini image models are part of the multimodal Gemini family, meaning image generation is now handled by the same model infrastructure that processes text and code.
What developers need to check
Before migrating, audit your codebase for these dependencies:
Model IDs in code
Search your codebase for any reference to Imagen 4 model strings. Common patterns include:
imagen-4-standardimagen-4-ultraimagen-4-fastimagegeneration@006(older Vertex AI naming)
These strings may appear in configuration files, environment variables, SDK initialization code, or hardcoded API calls.
SDK calls
If you use the Google AI SDK or Vertex AI SDK, check which version you are running. Older SDK versions may not support the new Gemini image model IDs. You will likely need to update your SDK to a version released after the Gemini image models became available.
Production dependencies
Identify every service, pipeline, or workflow that depends on Imagen 4 output. This includes:
- Backend services that generate images on demand
- Batch processing pipelines that create images in bulk
- Internal tools that use Imagen for content creation
- Third-party integrations that proxy through your Imagen calls
Each of these needs to be updated and tested before August 17.
Migration steps
Step 1: Change the model string
The most basic migration step is updating the model identifier in your API calls. Replace your Imagen 4 model ID with the corresponding Gemini image model ID.
Check Google’s official migration guide at firebase.google.com/docs/ai-logic/imagen-models-migration for the exact model string mappings, as these may change between now and the shutdown date.
Step 2: Update your SDK
Install the latest version of whichever Google SDK you use:
- Google AI Python SDK - Update to the latest version that includes Gemini image model support
- Google AI Node.js SDK - Same requirement
- Vertex AI SDK - Ensure your version supports the new model IDs
The API surface may differ between Imagen 4 and Gemini image models. Review the SDK changelog for breaking changes in how image generation requests are structured.
Step 3: Update request parameters
The Gemini image models may accept different parameters than Imagen 4. Review the API documentation for:
- Supported image sizes and aspect ratios
- Available style parameters
- Prompt formatting requirements
- Safety filter configurations
- Output format options
Do not assume that every parameter you used with Imagen 4 has an equivalent in the Gemini image API.
Step 4: Test output quality
Run your existing prompts through the new model and compare results. Image generation models can produce significantly different outputs even with identical prompts. Pay attention to:
- Style consistency with your existing image library
- Text rendering quality (if your use case involves text in images)
- Color accuracy and saturation
- Fine detail rendering
- Edge cases that your application depends on
Step 5: Update monitoring and error handling
Update any monitoring that references Imagen 4 model names, and ensure your error handling covers new error codes or response formats from the Gemini image API.
What is NOT a drop-in replacement
Be realistic about this migration. Google has not positioned this as a seamless swap, and developers should not treat it as one.
Image quality and style may differ. The Gemini image models are architecturally different from Imagen 4. Even with identical prompts, expect differences in artistic style, detail rendering, and overall aesthetic. If your application has strict visual consistency requirements, budget time for prompt engineering and quality testing.
API parameters may differ. Not every Imagen 4 parameter maps directly to a Gemini image model parameter. Some features may be missing, renamed, or work differently. Read the API reference carefully rather than assuming compatibility.
Pricing may differ. Google has not committed to price parity between Imagen 4 and the Gemini image models. Review current pricing at the time of migration and update your cost projections accordingly. For a broader view of image generation API costs, see our AI image generation API pricing comparison.
Response format may differ. The structure of API responses (image encoding, metadata fields, error formats) may differ between Imagen 4 and Gemini image models. Test your response parsing code thoroughly.
Rate limits may differ. Your existing Imagen 4 quota does not necessarily transfer to the Gemini image models. Check your quota allocations in the Google Cloud console.
Frequently Asked Questions
When exactly do Imagen 4 endpoints stop working?
August 17, 2026. After this date, all API calls to Imagen 4 Standard, Ultra, and Fast will fail. There is no extension or grace period announced.
Do I need to migrate if I use Imagen through a third-party service?
If a third-party service wraps Imagen 4 for you, contact that service to understand their migration plan. They may handle the switch transparently, or they may require you to update your integration.
Can I use both Imagen 4 and Gemini image models during the migration period?
Yes. Until August 17, both model families are available. This gives you time to run A/B tests, compare outputs, and gradually shift traffic to the new models.
Will my existing Imagen 4 generated images be deleted?
No. Previously generated images stored in your own infrastructure are not affected. The shutdown only affects the ability to make new generation requests.
Is there a way to keep using Imagen 4 after August 17?
No. Google has not announced any legacy access program or paid extension for Imagen 4. The shutdown is final across all tiers and access levels.
Sources
- Imagen models documentation - Official Google AI developer documentation
- Imagen models migration guide - Firebase/Google migration instructions
- Best AI API providers in 2026 - Comparison of alternatives if you want to evaluate other providers
Last updated: August 7, 2026. If you are evaluating alternative image generation providers during this migration, see our complete guide to AI API providers for options beyond Google’s ecosystem.