Google shut down the Imagen 4 API; the swap costs up to 67% more

The same Aug 17 purge retired Veo 3.0 and 2.0 video too, and both replacements are preview-tier. Plus: Cursor's cloud agents now boot 3x faster.

Nowline AUG 18 9:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • The three Imagen 4 IDs now return errors

    As of Aug 17, imagen-4.0-generate-001, its -ultra, and its -fast variants are retired in the Gemini API, so calls to them now fail. Google's pointed replacement is gemini-3.1-flash-image (aka Nano Banana 2).

  • Your image bill can jump ~67%

    Imagen 4 was flat-rate: $0.02/$0.04/$0.06 per image for fast/standard/ultra. The replacement prices by resolution instead, from $0.045 at 0.5K up to $0.151 at 4K, so a standard 1K image runs $0.067, roughly 67% over Imagen's $0.04 tier. Batch mode roughly halves it.

  • Veo 3.0 and 2.0 video went dark the same day

    The Aug 17 sweep also retired veo-3.0-generate-001, veo-3.0-fast-generate-001, and veo-2.0-generate-001. Google routes you to veo-3.1-generate-preview and its fast variant, but note those are preview, not GA.

  • The catch: your new media models are preview-tier

    gemini-3.1-flash-image is GA, but the Veo 3.1 replacements carry no GA stability guarantee and can shift or deprecate on short notice. Before you re-point production, re-test aspect ratios, quotas, latency, and the mandatory SynthID watermark.

  • Elsewhere: Cursor's cloud agents boot 3x faster

    Cursor's new Builds pre-warms dev environments in the background so Cloud Agents start up to 3x faster, and it keeps the last successful build when an environment fails. Live now on paid plans.