Prompts and images

Mynth sits in front of other companies' models. You send a model and a prompt, Mynth routes and retries the request, and the model itself runs at a provider.

Mynth does not review your content#

Mynth does not read your prompts or images, train on them, or use them for anything beyond the request you sent. There is no Mynth moderation pass on the way through.

Any other use needs your written approval. The terms and the privacy policy are the full rules, including the narrow case where the law requires us to act on something we have already become aware of. That is not a scan of your traffic.

Features you turn on read what they need, inside the request you made:

What the provider receives#

The provider sees the call as coming from Mynth. It does not receive your user id, your email, your API key, or the task's metadata.

It receives what the model needs: the prompt, any input images, the size, and the other generation fields that model accepts. When an attempt fails and Mynth retries on another provider, that provider receives the same fields.

Providers apply their own rules. Logging, retention, and content filters are theirs and differ between providers.

When the provider refuses#

RESTRICTED_CONTENT means the provider or the model refused the prompt or an input image. Mynth did not make that decision.

Mynth does not retry a refusal, because another attempt usually gets the same answer. Nothing stops you from sending the request again yourself, and it sometimes goes through.

On image.generate and video.generate, a refusal usually fails one output while the task still ends completed. The refused output is not charged. Tasks shows where the code sits on the result. The CLI exits with code 5 for it.

Next steps#

• Generate images: the request body.
• Errors: RESTRICTED_CONTENT and the other task codes.
• Privacy policy: prompts, images, and third-party providers.