Imagen 4 shut down Aug 17: Nano Banana 2.1 is the named replacement
Google lists gemini-nano-banana-2.1 as the Imagen 4 replacement. Imagen rows on Sume take no reference photos, while Nano Banana 2.1 takes up to 10.
Google lists three Imagen 4 ids with a shutdown date of August 17, 2026 and names gemini-nano-banana-2.1 as the replacement. That is a change in shape, not only in name: the Imagen rows in the Sume catalog are text-to-image only, while google/nano-banana-2.1 also takes reference photos for editing.
What Google says
The Gemini API deprecations page (fetched 2026-10-07) shows imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001 with an August 17, 2026 shutdown and the recommended replacement gemini-nano-banana-2.1. The release notes date Nano Banana 2.1 to October 6, 2026 and list 1K, 2K and 4K generation plus the ratios 1:4, 4:1, 1:8 and 8:1.
What differs on Sume
Sume lists google/imagen-4-fast, google/imagen-4-ultra and google/nano-banana-2.1 as separate catalog rows. Check GET /v1/images/models to see which of them your account can call today, since a vendor shutdown can outrun a catalog row.
| Capability | Imagen 4 Fast / Ultra | Nano Banana 2.1 |
|---|---|---|
| Reference photos | None, text-to-image only | Up to 10 |
| Resolution tiers | Ultra: 1K, 2K | 512, 1K, 2K, 4K |
| Ratio auto | No | Yes, auto is listed |
| Images per call | 1 to 4 | 1 to 4 |
Migration steps
A prompt tuned for Imagen will not match Nano Banana 2.1 output pixel for pixel. Run the same 10 prompts through both rows and compare before switching traffic.
- Replace the model id with google/nano-banana-2.1.
- Move any size logic to aspect_ratio plus resolution.
- Add references only if you want editing; Imagen never accepted them.
- Log usage.cost per call, because the two rows are priced differently.
Sources
Related posts
More in Models
- Lip-sync and motion control cost per clip: H3 Max, Fabric, Kling
Sume prices H3 Max lip-sync at $0.0625 to $0.20 per audio second, Fabric at $0.1875 and Kling 3.0 Motion Control at $0.1575. A 10-second clip is $0.63 to $2.00.
- Music Router prompt limit: 5,000 characters, and what to put in them
Sume Music Router prompts run 1 to 5,000 characters, with no duration field or negative prompt; length goes in the text. A brief budget and a checker.
- MAI-Transcribe-2-Streaming 0.13 s to final: when the clock starts
The 0.13 second figure is measured from end of speech found by a VAD, and partials arrive in about 100 ms. Why neither number is the wait for a file transcript.
- MAI-Transcribe-2-Streaming 2.5% WER: what the test audio mix is
The 2.5% word error rate comes from a chunked-streaming index: 50% AA-AgentTalk, 25% VoxPopuli, 25% Earnings22. Here is how that maps to a recorded call.
Written by Sume