Why did my generation fail?
The card tells you the reason — the most common are a content-policy refusal, a reference clip longer than the model allows, an audio track that is too long, a missing reference, or a model that does not accept photos of real people — and a failed generation is never charged.
Content policy
The provider refused the request. Prompts and reference images are moderated by each model provider, and the rules differ between them. Named celebrities, brands and titles, sexual or violent content and a few other categories are refused by most models. Rewrite the prompt — describe the look instead of naming the person — rather than retrying the same text.
Reference too long
Every video-input model has a maximum clip length, listed on the model page. The Workspace measures your file before generating and refuses one that is over; if a limit is hit at the provider instead, the card says so. Trim the clip and try again.
Audio too long
Lip-sync and audio-driven models cap the audio length, and the price follows the audio. Cut the track to the model's maximum.
Missing reference
"This model needs at least one reference image" means the row cannot run on text alone. Add the input the slot asks for.
Real people in references
Some models refuse a reference that shows a recognisable real person. Use a generated character, or a model that allows it.
Provider errors
Occasionally the provider fails or times out. LUVI recovers what it can, marks the rest as failed, and charges nothing. Retry after a minute; if it keeps happening on one model, tell us the generation link.
Frequently asked
Was I charged for the failed run?
No. Credits are only charged when a result is delivered.
Can I retry the same prompt?
For a content-policy refusal, rewrite the prompt first; the same text gives the same answer. For a limit error, fix the input; for a provider outage, wait a minute and retry.
Where do I see the error?
On the generation card in the Workspace and in the Library; the message names the cause.