Troubleshooting

Check the address, credentials, model and balance before diagnosing an upstream failure.

HTTP status checklist#

Status	Check first
`400`	JSON, required fields, parameter ranges and media references
`401`	Key validity and the `Bearer` header
`403`	Permissions, source IP, model scope and group
`404`	Endpoint, model/task ID, task ownership and duplicate `/v1/v1`
`429`	Rate, concurrency, token limits or upstream throttling
`5xx`	Platform/upstream availability, channel connection and timeouts

Insufficient balance can appear with different HTTP statuses or business codes. Read the response body too. SSE errors may occur after streaming starts; an initial 200 is not enough to establish success.

Catalog model unavailable to a key#

Check the group, model restrictions, expiry and quota. Use the full API model ID rather than a translated display name.

Output takes too long#

Check streaming configuration and buffering in your client or proxy. Long contexts and reasoning can increase first-token latency. Use a reasonable timeout, retain request IDs and avoid endless retries.

Reference download fails#

Check public access, expiry, permissions and file limits. A link opening in your logged-in browser does not prove upstream accessibility. A sharing page is not necessarily a raw file URL.

No video in the submission response#

After receiving a task ID, follow Video tasks. Keep that ID and query progress; do not resubmit generation to refresh a task.

Information for support#

Time and timezone, request ID, order ID or task ID.
Model, endpoint, HTTP status and error code.
Redacted parameters and screenshots.

Never send passwords, verification codes, complete API keys, cloud secrets or still-valid signed URLs to private media.