Google says Imagen is shut down: test your Imagen call on Sume now

Google's Imagen page, read 2026-10-08, says "Imagen models are shut down" and are no longer available in the Gemini API, and it points to Nano Banana. If your code calls Imagen through Sume, the Gemini shutdown is not the test: Sume's own catalog is. Run one catalog call and one image request to find out.

Two checks

First list the catalog and look for the Imagen rows. Then send one small request to the row you use. Imagen 4 Fast costs $0.025 per image on Sume, so the probe costs 2.5 cents.

curl -s https://api.sume.com/v1/images/models \
  -H "Authorization: Bearer $SUME_API_KEY" | jq -r '.data[].id' | grep imagen

curl -s https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"google/imagen-4-fast","prompt":"a red apple on a white table"}' \
  -w '\nHTTP %{http_code}\n'

Reading the result

A 200 with a data[0].url means the row works today. A 400 or 404 on the model means it left the catalog. A 202 means the job is queued; poll it as described in Jobs and results.

Plan the move anyway

Because the upstream product is shut down, treat the Sume Imagen rows as something that may change. The replacement for text-only work is google/nano-banana-2.1 at $0.075 at 0.5K, which also edits. Move ids in config, not in code, so the switch is one line.

What to do with the answer

If the probe returns 200, keep using the row but add the fallback. If it returns an error, move the pinned id to google/nano-banana-2.1 and rerun your saved prompts to see which need rewording, because the two models do not read prompts identically.

Budget the move. If you produce 500 Imagen 4 Fast images a month at $0.025, that is $12.50. The same 500 on Nano Banana 2.1 at 0.5K cost 500 x $0.075 = $37.50, and at 1K $50.00. The move triples or quadruples the bill, so decide whether the extra capability is worth it.

Watch your error rate rather than the upstream news. A single catalog call at startup is enough to know whether the row is listed.

Why not wait for an error

An error will arrive during a job, usually at the worst moment. A probe at deploy time costs $0.025 and tells you the state before the batch starts. Add it to your health check, but run it rarely, since each run bills an image. A catalog read is free of image charges and can run more often.

Keep the catalog read as the primary check and the paid probe as a secondary check.

If you run a small agency, put the probe in a weekly job rather than a deploy step. One probe a week is 52 x $0.025 = $1.30 a year, which is trivial next to the cost of a failed client batch. Pair it with a catalog read per deploy, which has no image charge, and alert when either one changes. That gives early notice at a known, tiny cost.