imagine-cinematic-scenes

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)

• Post: https://x.com/imagine/status/2092675531057885288
• Article: https://x.com/i/article/2092302666345164800

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

1. Speed Mode to explore character, location, aesthetic.
2. Quality Mode to upscale the keeper (often 720p Speed → 1080p Quality).
3. If a tint or stock must survive the upscale, say it in the Quality prompt.
4. 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:

1. What is the shot like? (size, angle, move, lens)
2. What are they doing? (action, eyeline, not looking into camera unless scripted)
3. How is the speech? (beat, pause, emphasis, pace). Dialogue only when a mouth is open.
4. 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.