Google’s older Vertex AI media-generation endpoints are approaching a hard deadline. Developers using several Imagen image-generation model IDs or Veo video-generation model IDs need to migrate before June 30, 2026, or risk broken creative workflows, failed API calls, and production jobs that depend on retired model names.
The cutoff is easy to miss because it sits inside model lifecycle documentation rather than a splashy product launch. But it matters for teams that built image or video generation into marketing tools, creative automation pipelines, internal content systems, agent workflows, or apps that call Google’s generative media APIs directly.
Google’s Vertex AI release notes say several Imagen generation endpoints are deprecated and should be updated before June 30 “to avoid service disruption.” The same March 24 notice lists older Veo GA endpoints with a June 30 migration target. Google’s Gemini API release notes, updated June 15, separately warn that Veo 2.0 and Veo 3.0 generation models will be shut down on June 30.
Which Vertex AI and Gemini media endpoints are affected
For image generation on Vertex AI, Google’s migration table points older Imagen and image-generation endpoints to gemini-2.5-flash-image. The discontinued image endpoints listed in the Vertex AI release notes include imagegeneration@002, imagegeneration@003, imagegeneration@004, imagegeneration@005, imagegeneration@006, imagetext@001, imagen-3.0-capability-001, imagen-3.0-capability-002, imagen-3.0-fast-generate-001, imagen-3.0-generate-001, imagen-3.0-generate-002, imagen-4.0-fast-generate-001, imagen-4.0-generate-001, and imagen-4.0-ultra-generate-001.
For video generation on Vertex AI, Google lists veo-3.0-generate-001, veo-3.0-fast-generate-001, and veo-2.0-generate-001 as discontinued endpoints, with veo-3.1-generate-001 or veo-3.1-fast-generate-001 as the recommended replacements depending on the old model used.
The Gemini API documentation adds a wrinkle for developers who are calling Gemini-facing endpoints rather than Vertex AI directly. Its June 15 changelog says veo-2.0-generate-001, veo-3.0-generate-001, and veo-3.0-fast-generate-001 are being shut down on June 30, and tells developers to move to Veo 3.1 preview model IDs or the GA models available through Gemini Enterprise Agent Platform.
Why this is more than a find-and-replace job
For simple prototypes, the migration may look like swapping one model string for another. Production systems usually need a little more care. Image and video models can differ in prompt sensitivity, default output style, latency, pricing, safety filtering, aspect-ratio handling, region availability, response schemas, and long-running operation behavior. A workflow that technically succeeds after a model-name change can still produce assets that fail brand review, exceed budget expectations, or break downstream processing.
Teams should start by searching their codebase, notebooks, workflow builders, prompt-management systems, and environment variables for the old model IDs. The risky places are not always obvious. A retired model name can be buried in a marketing automation template, a serverless function, a Zapier-style integration, a CI job that generates demo assets, or a low-code internal tool owned by a non-engineering team.
It is also worth checking vendor-managed integrations. If a product uses Google’s generative media APIs behind the scenes, the owner of that product may handle the migration. If an organization supplied its own Google Cloud project, API credentials, region, or model ID, the responsibility may sit with the customer instead.
A practical migration checklist before June 30
- Inventory model IDs. Search for
imagegeneration@,imagetext@001,imagen-3.0,imagen-4.0,veo-2.0, andveo-3.0across application code, notebooks, job configs, prompt tools, and secrets-backed settings. - Map each call to the recommended replacement. Vertex AI image-generation calls generally move to
gemini-2.5-flash-imageunder Google’s March 24 table. Vertex AI video-generation calls move from older Veo IDs to Veo 3.1 GA IDs. Gemini API Veo calls should follow the June 15 Gemini deprecation guidance. - Retest prompts, not just API success. Save representative prompts and compare output quality, safety behavior, file formats, aspect ratios, latency, and failure modes before changing production routing.
- Check downstream assumptions. Creative pipelines often expect a particular image size, video duration, storage path, metadata field, polling pattern, or moderation response. Those assumptions should be tested with the replacement model.
- Update monitoring. Add temporary alerts for model-not-found errors, operation failures, quota spikes, and unusual latency after the migration. The first sign of a missed model string may be a failed scheduled content job.
- Leave a rollback plan that does not depend on the retired endpoint. After shutdown, rolling back to the old model ID will not help. Safer fallback options include routing to an alternate supported model, pausing automated generation, or requiring manual review until output quality is stable.
The Vertex AI name change can add confusion
Google’s product naming also complicates the migration. The Vertex AI documentation now notes that Vertex AI services are part of Gemini Enterprise Agent Platform, Google Cloud’s platform for building, deploying, and governing enterprise AI agents. Some documentation, console surfaces, and third-party tutorials still use Vertex AI naming, while newer materials point developers toward Agent Platform.
That does not mean every Vertex AI system disappears on June 30. The immediate deadline is narrower: specific generative media model IDs are being retired. Developers should avoid turning this into a broad platform migration panic, but they should treat any hard-coded Imagen or Veo endpoint as technical debt with a near-term expiration date.
Who should act now
The most exposed teams are those that generate images or videos automatically: ad creative systems, ecommerce content tools, social video workflows, internal design assistants, media localization pipelines, and AI agents that create assets as part of a larger task. These systems often run on schedules, process queues, or campaign deadlines, which means a June 30 failure may show up after a team has stopped actively watching the integration.
Developers who only experimented with Imagen or Veo in a notebook may be able to update quickly. Teams with approval workflows, brand constraints, agency handoffs, or compliance review should move sooner. In those environments, “the API still returns something” is not enough; the replacement model needs to produce assets that fit the same operational and review process.
The best short-term move is simple: audit model IDs this week, migrate one representative workflow first, and run output comparisons before changing every production route. June 30 is close enough that waiting for a failed job is a poor test plan.
