MIXED REFERENCES
Direct MiniMax H3 with image, video, and audio references
Use the official six-section MiniMax H3 reference format, stable media labels, and fixed retention relationship markers.
Full-reference MiniMax H3 prompts are structured rewrites, not loose lists of influences. Write the six official sections in English, preserve the original language only for dialogue, lyrics, and visible scene text, and keep every reference label stable from its definition through the timeline and audio sections.
Keep the six sections in order
Write subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape, and non_diegetic_music in that exact order. Definitions establish what each source means. The summary starts with a bracketed task type such as [reference generation + audio reference]. Retention records what is preserved or transferred, and the detailed description places every reference into timed shots.
Assign the correct stable label
Use <Subject N> for reusable visible content such as a person, object, environment, style, or action. Use <Picture N> only when an image is a concrete frame or composition anchor; <Video N> for editing, continuation, or whole-video structure; and <Audio N> for an audio signal that is copied or referenced. A visible subject taken from a video is still a Subject, while Video identifies the asset or temporal structure.
- <Subject N>: reusable visible content
- <Picture N>: concrete frame anchor
- <Video N>: source or temporal structure
- <Audio N>: copied or referenced signal
Use the fixed retention markers
For Subject, Picture, and Video relationships, use only fully_preserved, partially_preserved, attribute_transfer, or weak_reference. For Audio, use fully_copy, partially_copy, reference, or weak_reference. Give each label one line that states where it appears and explains the relationship. Do not treat newly requested actions or backgrounds as losses when they do not contradict the role you defined.
Place references in the playback timeline
In detailed_description, establish style before [Shot 1], then specify composition, appearance, position, environment, light, actions, camera, sounds, and the exact moments when references take effect. Reuse stable (S1) speaker IDs and <d>[Language] ...</d> dialogue. Minimax Motion accepts up to 9 images, 3 videos, and 3 audio files, with separate 15-second total budgets for video and audio; start with the smallest set that has clear roles.
Reference workflow checklist
- Write all six sections in the official order
- Keep Subject, Picture, Video, and Audio labels stable
- Use only the fixed retention markers
- Describe exact reference use shot by shot
- Stay within the workspace media limits
Put the workflow into motion
Choose Standard or References, set the duration and resolution, and see the exact credit cost before you generate.
Create with MiniMax H3