| name | Imagine cinematic scenes |
| description | Use when directing Grok Imagine scenes or Odyssey-style cinematic video: official @imagine shot language, Speed then Quality stills, omni-reference video, and director-style prompts. Not for Phoenix Tiberian Sun plates. |
Imagine cinematic scenes
Use this when generating Grok Imagine stills or video that must read as a film scene, especially Odyssey / contest work.
Source of truth: official @imagine guide (2026-08-26)
Do not
- Prompt the word "cinematic" as the look. It is too general.
- Ask Imagine for a whole movie in one generation.
- Start video before stills/refs are approved.
- Skip the aim question.
1. Aim first
Answer: what should this scene do for the viewer? Shot list follows that. Same story can be cunning, grief, or power. Pick one.
2. Prompt with film words
Shot size: extreme wide, wide, medium, close-up, extreme close-up, over-the-shoulder, POV.
Angle: eye level, low angle, high angle, bird's-eye, dutch angle.
Move: camera static, pan, tilt, dolly in, dolly out, push-in, pull-out, handheld.
Lens: 24mm, 35mm, 50mm, 85mm, shallow depth of field.
On canvas, you can generate several still sizes in one go, then pick.
3. Still workflow
- Speed Mode to explore character, location, aesthetic.
- Quality Mode to upscale the keeper (often 720p Speed → 1080p Quality).
- If a tint or stock must survive the upscale, say it in the Quality prompt.
- Lock one face/location still before video.
4. Video workflow
Use omni-reference video (multiple images). It is the smartest current model for consistent characters, locations, and props.
- Attach the approved stills (≤7).
- Mark whether the image is a first frame or a guide.
- One shot = one generation. Typical 6–12s, 15s max.
- After a setup clip, later clips start already in motion.
5. Prompt like a director
Long prompts need all four:
- What is the shot like? (size, angle, move, lens)
- What are they doing? (action, eyeline, not looking into camera unless scripted)
- How is the speech? (beat, pause, emphasis, pace). Dialogue only when a mouth is open.
- What else to remember? (skin texture, env audio, no music unless the joke is a song)
6. After the clip
Check frames at ~2s, ~8s, ~13s, and last. Reject face/costume/location drift. Stitch in Imagine canvas only after keepers pass.
If a local production pack exists, follow its shot list and attach list exactly. Do not invent scenes.
