For Seedance video requests, a missing first frame in
image_first_frame mode returns
provider.invalid_request with error_params.field set to assets.firstFrameImage.
Unknown keys inside assets return HTTP 400. Duration errors include the supported
range for the selected model and mode. If a queued submission is rejected, retrieve
the task and read failureReason and, when available, error.upstreamCode.
Wan 3.0 and Prime preserve upstream rejection status, error code, message and
supplementary details, with credentials redacted. Rate-limit rejections are returned
without automatic queuing or retry. Failed task details appear in failureReason.
Retrieving an older failed task may recover a more specific reason if the provider
still retains it. If the provider has no further details, the original failure remains.
Google image errors distinguish timeouts, connection failures, oversized or unreadable
responses, and responses without an image. When Google supplies a known content-filter
or output-token-limit reason, the error describes that cause. Existing error_code
values are preserved; error_params.providerReason can add the classification.
An explicit upstream rejection message still takes priority after credential redaction.
Common status codes
Retry safely
- Branch on
error_codeand HTTP status, not message wording. - Correct invalid parameters, unsupported media, or content-policy failures before submitting again.
- Retry read requests after transient
5xxor429responses with exponential backoff and jitter. - If generation returned a task
id, poll that task instead of creating another one. - A timeout or interrupted create request does not prove that generation failed. Repeating it can create a separate billable task. If the outcome is unknown, check your task history or contact support before resubmitting.
- A
404response does not reveal whether a resource exists in another workspace.
Task failures
A successful HTTP 200 task lookup can return a failed generation. Always inspectstatus
before reading outputs. Check the endpoint reference for its terminal status values.
BytePlus tasks and individual failed images can include error.code and error.message.
A partial image response may contain both successful images and errors. Other task APIs
can return failureReason and an optional structured error with code, upstreamCode,
and message. Use these details to correct the input or share diagnostics with support.
Detailed error references
- BytePlus errors:
provider.byteplus_*codes for video, image, and text requests. - Generation task errors:
provider.mulerouter_*codes for Flash VSR, Wan, z-image, Qwen Image Edit, and Face Swap. - OpenAI image errors: image request failures and diagnostic IDs.
error_params.upstreamCode is an optional diagnostic code. Use the GENGEN error_code
for application logic. Unknown failures may use provider.request_failed.
provider.mulerouter_insufficient_balance is a model service issue. Contact GENGEN
support; adding funds to your GENGEN wallet does not resolve this error. Top up only
when the response identifies insufficient GENGEN wallet balance.Contact support
Include the HTTP status,error_code, request time, model, public task ID, and any
diagnosticId or upstreamCode returned. Never include your API key. See the
FAQ for common media and content-policy failures.