Image generation API: what changes when you move from a UI to code

Most people arrive here from a web UI where you type a prompt and wait. Four things change when the same models are behind an API.

1. You get a job, not an image

{ "job_id": "abc123", "state": "queued", "credits_held": 4 }

Poll it, stream it, or receive a webhook. The upside is that nothing has to stay connected — the downside is that a failed job arrives as a successful HTTP response describing a failure, and your error handling will not catch it unless you check state.

2. Billing is per call, not per month

Model	Credits per image	Batch	Resolution
Flux	5	up to 4	480p, 720p, 1080p
Nano Banana	4	up to 4	480p, 720p, 1080p
Ideogram	6	up to 4	480p, 720p, 1080p

No subscription tier gates which model you can use. You hold a credit balance and each call draws down from it; failed jobs refund in full.

3. There are ceilings, and they are per key

Tier	Requests per minute	Concurrent jobs
Free	5	1
Pro	60	5

One in-flight job on the free tier. Exceeding either returns a 429 — and the two 429s need opposite responses, so branch on error.code rather than the status. See api rate limit exceeded and CONCURRENCY_LIMIT_EXCEEDED.

4. Parameters are validated locally

A value outside a model’s schema fails before anything is submitted, with the offending field named in error.param. No credits are touched. Each model has a different schema — see AI image API for the per-model references.

The minimal working call

curl -X POST https://genace.ai/api/v1/images/generations \
  -H "Authorization: Bearer $GENACE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"a red ceramic mug on a white table"}'

No model needed — it defaults to Nano Banana at 4 credits ($0.04 at the Enterprise rate).

Models covered on this page

• FLUX Pro
• Nano Banana
• Ideogram v3