The three fields for text and keyframe prompts
integrated_multimodal_description is the main body. It runs along the timeline and holds the visual style, composition, subjects, actions, camera moves, dialogue, singing and any sound the characters can hear. Start [Shot 1] with the overall style and the opening composition, for example "Live-action, cinematic, a medium-wide shot frames…".
overall_soundscape summarizes ambient sound, physical action sounds and non-verbal human sounds in one to four sentences. Do not repeat dialogue here. MiniMax's guide reserves N/A for a video that should be completely silent.
non_diegetic_music describes only the score the audience hears and the characters do not, in one to three sentences about instruments, tempo and dynamics. Write N/A when there is no score.
integrated_multimodal_description: [Shot 1] <style>, <shot size> frames <subject> in <place>. <Action>. The camera <move>… [Shot 2] At 00:05.000, the camera cuts to <new information>… overall_soundscape: <ambience, action sounds, breathing, footsteps> non_diegetic_music: <instruments, tempo, dynamics> or N/A
Shots, cuts and camera moves
Number shots [Shot 1], [Shot 2] and so on. The first shot has no timestamp; every later shot opens with a single, strictly increasing cut time inside the clip, such as "At 00:03.500, the camera cuts to…". Cut when the subject, space, state, viewpoint or time changes. If only the framing changes a little, use a camera move instead.
Write a camera move as a natural sentence with a type, and an amplitude and speed when they matter: "The camera pushes in with small amplitude at slow speed toward the letter in her hands." The guide's move types include push in and pull out, pan, truck, tilt, pedestal, arc, tracking, static, shake, POV and roll.
Dialogue and on-screen text
Give every voice a stable speaker ID such as (S1) or (S2), with a short description of the voice the first time it appears. MiniMax's guide wraps the spoken words in a dialogue tag with a language label, for example: The young woman with a quiet, breathy voice (S1) says: <d>[English] I get off at the next station.</d>
Put visible text such as signs and labels in double quotation marks, verbatim and in its original language: A red neon sign reading "OPEN" glows above the doorway. Write the rest of the prompt in English.
First & Last Frame mode
With only a first frame, describe the image as the starting point and move forward: first-frame anchor, the action that starts, how it develops, the result. With a first and a last frame, describe the path between them rather than the two still images, and reach the last frame at the end of the final shot. MiniMax's guide prefers a single shot for first-and-last-frame clips unless you need cuts.
The guide also opens keyframe prompts with an alignment line, followed by a blank line and the three fields. For a first frame only it reads: For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
Omni Reference mode
For prompts built on reference files, MiniMax's guide uses six sections in a fixed order: subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape and non_diegetic_music. Give each reference a label and use the same label in every section. In this workspace, Omni Reference takes up to nine images, three video clips and three audio files, twelve files at most, and an audio file needs at least one image or video alongside it.
Example: a two-shot text-to-video prompt (10 seconds)
Choose Text to Video and 10 seconds. The cut lands at the five-second mark, so each shot gets about half of the clip.
integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium-wide shot frames a fisherman in a yellow raincoat coiling a wet rope on a wooden harbour pier at dusk. The camera pushes in with small amplitude at slow speed as he loops the rope over his forearm and glances at the darkening sea. [Shot 2] At 00:05.000, the camera cuts to a close-up of his weathered hands tying the rope to a metal cleat, rainwater dripping from his sleeves as he pulls the knot tight. overall_soundscape: Steady rain patters on the wooden planks while small waves slap against the pier posts. The wet rope creaks against the metal cleat and gulls call in the distance. non_diegetic_music: N/A
Example: a first-frame prompt (5 seconds)
Choose First & Last Frame, upload a still of a person at a café window, and set 5 seconds.
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced. integrated_multimodal_description: [Shot 1] Live-action, cinematic, the man shown in <Picture 1> stays seated at the rain-streaked café window, keeping his appearance, clothing, seat and the table layout. The camera trucks right with small amplitude at slow speed as he lifts a white coffee cup, takes a slow sip and sets it back on the saucer, then turns his head toward the street. overall_soundscape: Rain ticks against the glass over low café chatter. The cup clinks softly on the saucer and a chair scrapes somewhere behind him. non_diegetic_music: Sparse piano notes at a slow tempo with a soft, sustained low string underneath.