AI Launch Profile
Google Imagen 4 API shutdown window: August 17, 2026
Google announced that imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001 are deprecated. Its deprecations table lists August 17, 2026 as the earliest possible shutdown date and recommends gemini-3.1-flash-image.
At a glance
Launch Snapshot
- Company
- Launch date
- June 15, 2026
- Launch type
- Not classified
- Category
- AI Developer Tools, AI Image Tools, AI Models
- Audience
- AI Artists, AI Platform Teams, Developers, Product Teams
- Pricing
- There is no separate migration fee. Replacement Gemini API image generation follows current Google pricing and quota terms; teams should compare the selected model, image output, prompts and workload mix on the live pricing page.
- Free plan
- No
- API
- Yes
- Open weights/source
- No
Launch Context
Use these links to move from this record into the broader Launch Intelligence database.
Verification & Sources
- Status
- Verified
- Source links
- 5
- Freshness
- Verified July 28, 2026
- Last verified
- July 28, 2026
- Last updated
- July 28, 2026
Key source checks
Suggest a correction
Kingy AI Take
Plan against August 17 while preserving Google’s wording: the deprecations page says listed dates are the earliest possible shutdown dates and that exact timing will be communicated in advance. Move a representative test suite to gemini-3.1-flash-image now and inspect prompt adherence, aspect ratios, safety handling, SynthID implications, latency, quota and cost before routing production traffic.
Who it is for
Gemini API developers, creative-platform teams and product owners using any of the three stable Imagen 4 endpoints for image generation.
What feels promising
Google names a replacement, maintains an explicit deprecation table and documents current native image-generation behavior, so teams can test the new surface before the earliest shutdown window.
What feels unproven
Kingy did not compare Imagen 4 with gemini-3.1-flash-image. Visual quality, prompt behavior, safety responses, output controls, quota, latency and cost can differ and must be measured on the application’s own prompts.
Source-backed record