Choose base or full-reference mode first

MiniMax separates guidance for text or keyframe modes from full-reference Ref2VA work. Start by naming the production problem. A text-only brief needs a complete visible and audible plan. A first-frame or last-frame brief must explain how the scene moves away from or toward the supplied image. A full-reference brief must define the role of every image, video, and audio item rather than asking the model to infer why the files were attached.

• Text and keyframes: describe the whole target shot
• Full reference: label each source and relationship
• Use only assets you are authorized to process

Describe the shot on a timeline

Write the subject, setting, framing, lens feeling, camera behavior, action, performance, lighting, and soundscape in the order a viewer experiences them. For a multi-beat result, give meaningful timing anchors without forcing an edit every second. State when dialogue begins, when an action changes, and what motivates a camera move or cut. Keep the requested duration within the exact product surface rather than assuming the repository's documented envelope is exposed everywhere.

Assign every reference one job

Label an image as identity, product appearance, style, first frame, or last frame. Label a video as motion, camera, performance, pacing, or environment reference. Label audio as dialogue, voice character, music, ambience, or timing. When two sources disagree, state the priority. For example, preserve the product geometry from one image while borrowing only camera motion from a reference clip and only vocal timing from an authorized audio file.

Write acceptance criteria beside the prompt

Prompting is incomplete without review. Define the non-negotiable identity or product details, the required story beat, acceptable camera behavior, audio synchronization, text legibility, disclosure needs, and forbidden changes. After generation, revise the smallest failed decision and preserve the original prompt, references, settings, output, and reason for the change. This makes the creative process repeatable and prevents a polished but inaccurate clip from passing review.