Errors and Billing
A request that Ideogram rejects before it runs is not billed. A request that runs is billed when it completes, including when an image is withheld by the safety check. A generation that fails is not billed. Ideogram doesn’t automatically refund a completed request.
Outcomes and billing
The 422 row applies to v2 endpoints, which check the prompt before the request runs.
A timeout doesn’t stop the work
A synchronous request waits up to 265 seconds for its result. When the wait runs out, the request returns 503, but only your wait ends: the generation keeps running and is billed when it completes. The 503 response has no generation_id, so you can’t poll for that result, and sending the request again starts a second, separately billed generation.
For anything that can take long, send the request asynchronously (async or a webhook_url), record the generation_id, and poll or use a webhook. A 503 whose error says the endpoint or model is temporarily unavailable is different: nothing ran, so you can retry it. See Idempotency and safe retries.
Error bodies
Rejections for account and usage limits use this shape:
Read reject_reason rather than the error text, which is written for people and can change.
Check what you were charged
usage_cost_usd_microson a completed generation, when the request uses variable usage-based pricingGET /v2/account/usagefor usage over a date rangeGET /v2/account/invoicesfor invoices