Hailuo·von MiniMaxImage to video

Hailuo 02 i2v Pro

Hailuo 02 is a new AI video generation model from Hailuo AI.

Im Arbeitsbereich öffnenminimax/hailuo-02/i2v-pro

Parameter

ParameterTypStandardBereich oder Optionen
enable prompt expansion
enable_prompt_expansion

The model automatically optimizes incoming prompts to enhance output quality. This also activates the safety checker, which ensures content safety by detecting and filtering potential risks.

selectfalsetrue, false

Eingaben

Dateien, die dieses Modell zusätzlich zum Prompt entgegennimmt.

EingabeAkzeptiertMax. Dateien
image
image
image1Erforderlich

Prompt-Leitfaden für Hailuo

TARGET MODEL: Hailuo (MiniMax Hailuo-02 / Hailuo-2.3 / 2.3-Fast — the picture-only video line that preceded H3). Prose plus bracketed camera commands. No negative prompt, no weighting syntax, no seed; the model renders picture only — there is no dialogue, no sound design and no music to direct, so words spent on audio are wasted.

Regeln, die gelten müssen

  • Camera is a bracketed command from the closed list of fifteen, exact spelling, placed inline where the move happens: [Truck left] [Truck right] [Pan left] [Pan right] [Push in] [Pull out] [Pedestal up] [Pedestal down] [Tilt up] [Tilt down] [Zoom in] [Zoom out] [Shake] [Tracking shot] [Static shot]. Official form: "A man picks up a book [Pedestal up], then reads [Static shot]."
  • Commands inside ONE bracket run simultaneously — "[Pan left,Pedestal up]" — three at most. Sequential moves are separate brackets in text order ("… [Push in], then … [Pull out]"). Never invent a command (no [Dolly], [Orbit], [Crane], [Rack focus]); an unlisted move is written as plain prose, which works but binds less precisely.
  • Hard cap 2,000 characters; the sweet spot is one clear paragraph. One subject, one primary action, one setting, one camera treatment per clip — the model reads a short, concrete scene better than a list.
  • Describe motion physically (what moves, in which direction, how fast, what it touches) — Hailuo's strength is physics realism and it rewards verbs over adjectives. No praise words (beautiful, stunning, masterpiece, 8K).
  • Write exclusions as positive framing; there is no field for them and a bare "no X" is unreliable.
  • A real person's name, a film or a studio is not a valid subject; describe the look instead.

Was am besten funktioniert

  • Order: subject (appearance) → action → environment → light and mood → camera command(s) → style word. Style words the line responds to: cinematic, realistic film, anime, 3D animation, claymation, watercolor.
  • Hold the frame deliberately: "[Static shot]" when nothing should drift; "[Tracking shot]" to follow a moving subject; pair a move with its purpose ("[Push in] toward her hands as she opens the box").
  • Match the action to the clip: a 6-second clip holds one beat and one move; a 10-second clip holds two, joined by "then".
  • Fast tier and expansion are parameters, not words — never write "fast", "768P" or "6 seconds" into the prompt.

Aus LUVIs eigenem Handbuch für diese Familie; dieselben Regeln wendet der Rewriter im Arbeitsbereich und die MCP-Prompt-Engine an.

Weitere Modelle dieser Familie