Generations

Plan for free

POST /story_intelligence_report returns the proposed shot plan and the five advisory story checks. POST /director_prompt_plan returns the full research plan: cinematic idiom, timecoded edit beats, camera physics, depth layers, color intent, an artifact-risk score, and ready per-model prompts. Neither renders; neither costs.

generate_video

{
  "raw_intent": "15 second premium launch film for a ceramic kettle",
  "director_mode": "auto",
  "brand_id": "aurora-kettle",
  "shot_count": 4,
  "duration_seconds": 15,
  "model_override": null,
  "likeness_consent": false
}

raw_intent is required; everything else is optional. director_mode accepts any of the 14 mode ids or auto. shot_count accepts 3 to 5. model_override forces one model for every shot.

Response

{
  "status": "ok | blocked | render_failed",
  "mode_selection": { "primary": "tentpole", "secondary": "mythic_heroic", "reason": "..." },
  "shots": [
    {
      "shot_id": "shot_1",
      "role": "hook",
      "narrative_purpose": "...",
      "camera": { "lens_mm": 35, "movement": "slow_push_in", "angle": "slight_low" },
      "routed_model": "veo_3_1",
      "attached_ref_ids": ["ref_char_...", "ref_style_..."],
      "provenance": { "manifest_id": "...", "watermarked": true }
    }
  ],
  "story_report": { "checks": [ ... ] }
}

generate_video_staged

The look-lock flow: the anchor shot's full direction becomes an inspectable, provenance-signed hero-frame still from an image model, then that exact frame is locked as the continuity reference and animated by an image-to-video model. Both stages are policy-routed with failover.

animate_image

Bring your own image: animate_image_ref is locked as the continuity reference and the request routes to an image-to-video-strong model. Likeness in your image is subject to the same consent gate as text.