Google has scheduled three Imagen 4 endpoints in the Gemini Developer API for shutdown on August 17, 2026. If a marketing, creative-production or content workflow calls imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 or imagen-4.0-fast-generate-001, treat today as a migration deadline. Google’s lifecycle page says listed shutdown dates are the earliest possible dates, however, so this article does not claim that every endpoint was already switched off at the time of publication.
The immediate task is to find the affected model IDs, preserve the current output baseline, test a supported replacement and update fallbacks before the old endpoint begins returning errors. There is one complication: Google’s own pages disagree about the recommended replacement. The Imagen model page and deprecations table point to gemini-3.1-flash-image, while the Gemini API pricing page points to Gemini 2.5 Flash Image. That conflict should be recorded and resolved against the current account notice and live model documentation rather than hidden behind a generic “upgrade now” instruction.
Scope boundary: the official evidence here is about the Imagen 4 model IDs exposed through the Gemini Developer API. It does not establish the same shutdown state for Vertex AI, Google Ads creative tools, Google AI Studio’s interface, Gemini consumer apps or images that were already generated and stored elsewhere.
What Google has scheduled for August 17
Google’s Gemini API release notes announced the deprecation on June 15. The current Gemini deprecations table, last updated August 13, lists the same three stable Imagen 4 IDs with an August 17 shutdown date.
| Model ID | Lifecycle date shown by Google | Action for the workflow owner |
|---|---|---|
imagen-4.0-generate-001 | August 17, 2026 | Locate every direct call, test the supported replacement and update the production model setting. |
imagen-4.0-ultra-generate-001 | August 17, 2026 | Re-test high-fidelity and text-rendering jobs instead of assuming equivalent output. |
imagen-4.0-fast-generate-001 | August 17, 2026 | Re-test latency-sensitive batches, retry logic and cost controls. |
Google defines a shutdown as the point at which an endpoint is completely turned off and no longer available. The same page also says its dates are the earliest possible retirement dates and that users will receive advance notice of the exact date. Those two statements matter together: the August 17 entry is enough to justify urgent migration work, but it is not enough to assert that a particular account failed at a particular hour.
Why this matters to marketing and creative operations
An API model ID can sit several layers below the tool a marketer sees. It may be stored in an automation platform, a serverless function, a spreadsheet script, a CMS image helper or a custom creative pipeline. The person publishing assets may never see “Imagen 4” in the interface. A shutdown can therefore appear as a generic generation error, an empty image response, a retry loop or a silent fallback to a different model.
For teams using image generation in content production, the risk is not only downtime. A replacement can change typography, composition, aspect-ratio behavior, prompt interpretation, safety refusals, latency and the visual consistency of a campaign. DMT’s Grok Imagine Image 2.0 review covers a different vendor and product, but it illustrates the right evaluation principle: model access, editing behavior and production quality must be tested separately. A model name alone does not prove that an existing creative brief will render the same way.
Google’s replacement guidance currently conflicts
The most current lifecycle-specific pages are aligned. The Imagen 4 model page says to migrate to Gemini 3.1 Flash Image to avoid service interruption. The deprecations table also names gemini-3.1-flash-image for all three Imagen 4 IDs.
But the Gemini Developer API pricing page currently displays a warning under Imagen 4 that recommends Gemini 2.5 Flash Image. That is a first-party documentation inconsistency, not evidence that one model is universally better or that both are interchangeable. It may reflect a stale page, a migration path for a different use case or an update that has not propagated consistently. Google does not explain the discrepancy on those pages.
The safest operational reading is:
- Use the model page and lifecycle table as the primary source for the currently named replacement.
- Check the exact deprecation notice shown in the affected Google account or project.
- Confirm that the replacement model is available in the intended API, region, tier and project before changing production code.
- Test output, latency, limits and price with the current official documentation; do not infer parity from the word “replacement.”
- Keep the documentation conflict in the migration ticket until Google’s pages agree or support confirms the correct route.
DMT’s Gemini 3.7 Flash guide for marketers owns a broader model-and-agent evaluation job. This migration is narrower: it concerns image endpoints and the behavior of a creative pipeline. The same evaluation discipline applies, but the model IDs and acceptance tests are different.
A practical migration checklist
1. Inventory every place the model ID can live
Search source code, environment-specific configuration, workflow builders, scheduled jobs, template repositories and deployment variables for all three exact strings. Include inactive projects and disaster-recovery paths. A fallback that still points to Imagen 4 can turn a controlled migration into a later outage.
Record the calling service, owner, project, environment, request volume and last successful call. Do not publish API keys, project identifiers or customer prompts in the migration record. The goal is an actionable inventory without creating a new security problem.
2. Preserve a representative output baseline
Select prompts that represent the work the system actually performs: product backgrounds, social crops, blog illustrations, ad concepts, text-heavy designs, brand colors, people, regulated claims and unusual aspect ratios. Preserve the prompt, relevant parameters, accepted Imagen 4 output and human review decision. Existing generated assets do not need to be regenerated merely because the endpoint lifecycle changed.
If the workflow supports editing or reference images, include those paths separately. Image generation and image editing can have different input requirements and failure modes. The AI content-marketing tools guide provides broader tool-selection context, but a migration test should be based on the pipeline’s accepted assets, not a generic feature list.
3. Test the replacement in a non-production path
Change the model only in a development or staging environment first. Run the representative prompt set and compare accepted-output rate, substantive edits, typography defects, prompt-policy refusals, latency, retry behavior and cost fields. Separate what Google documents from what the team observes in its own sample.
Do not use a single attractive image as the acceptance test. A creative pipeline needs repeatability across common briefs and known edge cases. If the replacement produces attractive but off-brand work, the migration is technically successful and operationally incomplete.
4. Recheck provenance, rights and disclosure
A model replacement does not automatically preserve the organization’s rights process. Keep the original prompt and source-asset record, verify that reference images are licensed for the intended use, and record which model produced the final asset. Recheck the current provider terms, watermark or provenance behavior and any disclosure requirement that applies to the channel.
This is adjacent to DMT’s authentic AI-content and provenance checklist. That article covers cross-platform content governance; this API migration needs the same traceability at the asset level. Do not claim that an invisible credential, watermark or label exists unless the exact generation route documents and preserves it.
5. Update fallbacks, alerts and rollback
Move the new model setting behind configuration rather than hard-coding it in several services. Update retries so a retired endpoint does not loop indefinitely. Add alerts for model-not-found responses, elevated failures, empty outputs and unexpected fallback use. Keep the last working configuration and the accepted baseline so the team can diagnose whether a problem comes from the model, the SDK, the prompt or the surrounding system.
For multi-step automations, log the model identifier beside each result. The AI agent harnesses and production-controls guide explains the wider need for bounded tools, traces and accepted-output definitions. Those controls are just as useful when the “agent” is a creative pipeline with a model call in the middle.
What this update does not establish
- It does not prove that all three endpoints had already stopped responding when this article was checked.
- It does not provide an account-specific shutdown hour or migration-completion status.
- It does not establish that Vertex AI follows the identical lifecycle, model ID or replacement path.
- It does not establish that Google AI Studio’s interface or a consumer Gemini image feature is unavailable.
- It does not prove equivalent quality, latency, safety behavior, pricing or quotas between Imagen 4 and either named replacement.
- It does not prove that stored Imagen 4 outputs become invalid; the lifecycle notice concerns API endpoints.
These boundaries are important because “Imagen 4 shutdown” is broader than the primary evidence. The precise claim is that Google’s Gemini Developer API documentation lists three Imagen 4 endpoints for shutdown no earlier than August 17 and recommends migration to a supported image model.
A same-day action plan
- Now: search for the three exact model IDs and identify production owners.
- Next: check the project’s deprecation notice and current supported-model documentation.
- Before switching: run representative prompts through the candidate replacement in staging.
- At release: change the configured model, update fallbacks and monitor errors and accepted-output rate.
- After release: reconcile the documentation conflict, pricing, rate limits, asset provenance and any manual review burden.
If the old endpoint still responds, use that time to finish the migration rather than treating availability as a deadline extension. If it no longer responds, restore the creative service with the smallest verified replacement path and keep output review in place until the new baseline is understood.
Frequently asked questions
Which Imagen 4 model IDs are affected?
Google lists imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001 in the Gemini API deprecations table.
Did Imagen 4 definitely shut down everywhere on August 17?
The official pages schedule the endpoints for August 17, but Google also says table dates are the earliest possible shutdown dates and that exact dates are communicated to users. This article therefore does not claim universal completion at a specific time.
Should I migrate to Gemini 3.1 Flash Image or Gemini 2.5 Flash Image?
Google’s model page and August 13 deprecations table recommend Gemini 3.1 Flash Image. The pricing page currently recommends Gemini 2.5 Flash Image. Use the lifecycle-specific guidance as the starting point, then verify the affected project’s notice, current model availability and requirements. Record the conflict rather than assuming the two routes are equivalent.
Does this affect previously generated images?
The cited lifecycle notice concerns API model endpoints. It does not say that image files already generated and stored by a customer will stop working. Their continued use still depends on the organization’s rights, storage and channel policies.
Should a marketer handle the migration alone?
The content or brand owner should define the accepted visual baseline, while the technical owner updates the API integration, fallback and monitoring. Rights, provenance and channel disclosure may require a separate review. Treat it as one operational change with named owners, not a silent model-string edit.
Sources and editorial boundary
- Google AI for Developers: Gemini API release notes — June 15 deprecation announcement and the three exact model IDs.
- Google AI for Developers: Imagen 4 model page — August 17 date and Gemini 3.1 Flash Image recommendation.
- Google AI for Developers: Gemini deprecations — lifecycle definition, earliest-date caveat, August 13 update and replacement table.
- Google AI for Developers: Gemini API pricing — conflicting Gemini 2.5 Flash Image recommendation under the Imagen 4 warning.
All material product claims above are limited to those first-party Google pages checked on August 17, 2026. DMT did not test an authenticated Imagen 4 project, call the affected endpoints or verify an account-specific notice. The inventory, testing, rollout and governance steps are editorial workflow guidance, not Google performance promises or proof of availability in a particular account.
Author note: Tayeeb Khan reviewed this as a practical migration checklist for marketing and creative operations, with confirmed documentation, observed documentation conflict and unknown account-level behavior kept separate.
