Skip to main content
A generation that fails and a generation that is refused are two different problems with two different fixes. Start by reading which one you got.

Which one happened

If it failed

Gems spent on a generation that fails are refunded. Run it again first; most failures are transient. If it fails twice on the same settings, change one thing at a time: drop the resolution a step, shorten the duration, or remove an attached reference. Then try a different model on the same prompt. A failure that repeats across models is worth reporting: send the prompt, the model, the settings, and roughly when it happened to mage@mage.space.

If it was refused

Mage’s rules

The content rules apply on every model and plan, and no amount of rephrasing gets around them. See what you can and cannot create.

A lab’s filter

GPT Image 2, Grok Image, Grok Video, Nano Banana 2, Veo 3.1 Lite, and Veo 3 are run by their labs, and each lab’s content rules apply to your prompts before Mage sees the result. A refusal from one of these comes from that lab’s filter, and Mage cannot change or override it. The same prompt may run on a Mage model.

Refusals that are neither

  • An imported custom model that will not load is a compatibility problem, not a content one. See importing custom models.
  • A model above your plan’s unlimited tier does not leave the picker. It switches from unlimited to Gems and keeps running, which is also what happens after a downgrade.

What not to do

Do not retry a refused prompt repeatedly with small variations. If you think a refusal was wrong, say so once, with the prompt and the model, at mage@mage.space.
Last modified on September 14, 2026