MiniMax H3 prompt cheatsheet
Phrase bank for MiniMax Hailuo H3 video prompts. Each row is a pasteable line for camera, framing, voice, face, dialogue, sound, light, or prompt structure.
This repo holds the spreadsheet. It does not contain model weights.
File: miniax-h3-cheatsheet-categorized.xlsx
How to use a row
Every content sheet uses the same five columns.
| Column | What to copy |
|---|---|
| Category | The family the phrase belongs to, such as Camera Motions or Vocal Tones. |
| Keyword or Element | The move, tone, or rule name. |
| Snippet or Description Template | The wording to paste into a prompt. Fill the blanks. |
| Usage Example | That wording inside a finished shot. |
| Source | A number. Look it up on the Source references sheet. |
Keep the official verb. Land the move on something in the shot: toward…, revealing…, or as…. One camera move per shot.
From Camera Moves:
The camera pushes in with small amplitude at slow speed toward her face.
Sheets
Table 1 is the full index (184 rows). The next eleven sheets are those same rows split by topic. Source references is the bibliography those rows cite.
| Sheet | Rows | Use it for |
|---|---|---|
| Table 1 | 184 | Every phrase in one list. |
| Camera Moves | 25 | Push, pull, zoom, track, truck, pan, tilt, pedestal, arc, roll, static, POV, shake. Amplitude is small or large. Speed is slow or fast. |
| Framing & Lens | 18 | Shot size, angle, 50mm and 85mm wording, and the face-safe phrase bank. |
| Shot Cuts & Editing | 6 | Cut lines and timestamps such as 00:03.000. Shot 1 has no timestamp. |
| Voice Library | 23 | Timbre lines when there is no reference audio: F1–F10, M1–M11, K1, K2. |
| Vocal Delivery | 12 | Pace and delivery of a spoken line. |
| Faces & Performance | 35 | Expressions and emotion changes on a readable face. |
| Dialogue & Languages | 21 | Speaker IDs (S1), <d> dialogue, native script, closed lips after a line, on-screen text. |
| Sound & Music | 12 | overall_soundscape, foley, score, and the dialogue-only house rule. |
| Lighting, Color & Look | 14 | Key light, grade, film grain, diffusion, wet surfaces, aspect ratio. |
| Reference Modes | 11 | Ref2VA retention markers, task-type prefixes, identity lock, and the I2VA / FL2VA / L2VA instruction lines. |
| Prompt Structure | 7 | The opening The target video is … sentence and label grammar. |
| Source references | 15 | The notes each Source number points to. |
Modes these phrases are written for
| Mode | The pictures are |
|---|---|
| T2VA | None. Text only. |
| I2VA | One image, and it is the exact first frame. |
| FL2VA | Two images: the exact first frame and the exact last frame. |
| L2VA | One image, and it is the exact last frame. |
| Ref2VA | Character, wardrobe, place, motion, or voice references. They are not a literal frame. |
A Ref2VA prompt has six sections, in this order:
subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape, non_diegetic_music.
summary starts with one task type in brackets, such as [reference generation]. Retention markers are fully_preserved, partially_preserved, attribute_transfer, and weak_reference. Audio markers are fully_copy, partially_copy, reference, and weak_reference.
Source notes
| # | Note |
|---|---|
| 1 | 11-style-80s-bollywood.md |
| 2 | 03-camera-motions.md |
| 3 | 08-t2va.md |
| 4 | 09-i2va-fl2va-l2va.md |
| 5 | 13-style-modern-series.md |
| 6 | 10-ref2va.md |
| 7 | 14-style-noir-thriller.md |
| 8 | 12-style-cinematic-modern.md |
| 9 | 16-preflight-checklist.md |
| 10 | 15-troubleshooting.md |
| 11 | 05-voice-library.md |
| 12 | 06-expressions-and-delivery.md |
| 13 | 02-shared-grammar.md |
| 14 | 04-framing-and-timing.md |
| 15 | 07-languages-and-on-screen-text.md |
License
MIT.