Image Generation Grok Aurora

Operator wants Grok to generate an image from a text prompt. They call an image-generation endpoint with model id pinned to an Aurora variant (or a Grok image-generation model per docs.x.ai [REQUIRES-VERIFICATION] on current endp…

Pin endpoint path and model id from docs.x.ai/api — treat both as [REQUIRES-VERIFICATION]. Send the prompt, choose response_format ('url' vs 'b64_json'), and set n (number of images). Authenticate with the same Bearer API key as chat completions.

End user prompt: 'photo of a recognizable celebrity in compromising position.' Agent forwards verbatim to image-gen.

Pre-screen prompts against xAI's Usage Policy (sexual content of identifiable persons, CSAM, non-consensual intimate imagery, copyright/likeness misuse). Refuse client-side before submitting. If the prompt passes the gate, log the decision with the rationale. Honor any AUP signal in the response.

Image-gen response may include a revised_prompt field (model rewrote the request for safety or clarity). Agent ignores it.

Surface revised_prompt to the operator (and optionally end-user) so the difference between requested and generated prompt is visible. Log revised_prompt for audit — meaningful for moderation review. Do not silently swap a substantially different prompt without disclosure.