H3 Studio

MiniMax-H3 · video + audio

H3 Studio
GPU
Mode
Images
Drop images here, paste, or browse
Reference image size
Reference videos
Drop videos here or browse
Converted to 24 fps and cut to the video length. “Use its sound” also gives the model the clip's audio (voice, music, effects) as its own <Audio N>.
Reference audio
Drop audio here or browse (voice, music, sound)
First 15 s is used. Up to 12 references in total.
Prompt0
Prompt tips
  • Write actions as processes: starting state → trigger → movement → reaction → end state.
  • Be concrete about the camera: position, direction, speed, and what it follows.
  • Tie effects to physics: sparks at contact, dust from footsteps, hair and cloth reacting.
  • Audio is generated too: describe ambience, synced sound effects, music, and dialogue as (S1) speaks: "…".
  • Multi-shot: [Shot 1] … then [Shot 2] At 00:03.000, …
  • Reference mode: refer to images as <Picture 1>, <Picture 2>…, videos as <Video 1>… and audio as <Audio 1>… Say what each one provides (identity, voice, motion, scene, style).
  • Avoid generic words like “cinematic” or “epic” — describe what is visible.
Aspect ratio
Resolution
Duration
Keyframe (optional)
Drop an image to lock the video to it at a chosen moment
Sampling
Sampler
Shift · video / audio
Official 12 / 3. Higher spends more steps on layout and motion, lower on fine detail. Only for Full · 20 steps (the Turbo LoRAs are trained for fixed settings).
LoRAs
Seed

Your videos