"Animate this still: [motion description]. The camera [holds steady / pushes in slowly]. Subject stays in frame."
Separate what MOVES from what the CAMERA does. H3 follows both when they are named apart.
MiniMax H3 — the step up from Hailuo 2.3 in the same family. Native 2K output and the strongest prompt adherence of the MiniMax models, with up to five reference images to carry a subject across shots. Clips run 4 to 15 seconds.
Video models start at Pro.
▶ 0:15
▶ 0:15
▶ 0:15
▶ 0:15
▶ 0:15
▶ 0:15
▶ 0:15"Animate this still: [motion description]. The camera [holds steady / pushes in slowly]. Subject stays in frame."
Separate what MOVES from what the CAMERA does. H3 follows both when they are named apart.
"Use the product in the reference images as the subject. [Scene and motion]."
Up to five references, and the first five cost nothing — this is the model to reach for when a subject has to survive across shots.
"[Scene]. — then set duration to 8-15s for a full beat rather than a loop."
Most models here stop at 6 to 8 seconds. H3 goes to 15, which is long enough for a shot to actually resolve.
Same family, different tier. H3 outputs native 2K where Hailuo 2.3 tops out at 1080p, runs to 15 seconds rather than 6, and takes up to five reference images. Hailuo 2.3 is the value pick; H3 is the one to use when the prompt is long or the subject has to stay consistent.
The provider builds a 4-second clip regardless of what you ask for below that, so a shorter request would pay for four seconds and receive one. The floor is enforced on our side instead.
When the prompt requests video output and the requested duration / resolution falls within MiniMax H3's capabilities.
Yes. The model picker in any chat lets you pin per-turn or per-session. Set as default in /settings.