Hailuo camera commands: the 15 bracket options like [Push in]
MiniMax's Hailuo API reads 15 camera commands in square brackets, such as [Truck left] or [Push in]. The full list, and what to do on Sume.
MiniMax's Hailuo video API reads 15 camera commands written in square brackets inside the prompt, for example [Truck left] or [Push in] (read 2026-10-02). The docs say several commands inside one bracket take effect at the same time. Sume has no camera-command parameter on its video route, so on Sume you describe camera movement in the prompt text instead.
What are the 15 commands?
MiniMax lists them on the text-to-video reference, and the image-to-video reference repeats that the newer Hailuo models support the bracket syntax (both read 2026-10-02). The names below are as printed there.
| Group | Commands |
|---|---|
| Truck | Truck left, Truck right |
| Pan | Pan left, Pan right |
| Dolly | Push in, Pull out |
| Pedestal | Pedestal up, Pedestal down |
| Tilt | Tilt up, Tilt down |
| Zoom | Zoom in, Zoom out |
| Other | Shake, Tracking shot, Static shot |
Does the syntax apply to H3?
The pages I read attach the bracket syntax to the Hailuo models (MiniMax-Hailuo-2.3 and 02) and say nothing about it for MiniMax-H3, so do not assume it carries over. If you need H3, MiniMax's own video guide describes the model's inputs and limits, and plain-language shot direction in the prompt is the safer start. Treat any claim that H3 parses brackets as untested until you try one clip each way.
What does Sume do with camera direction?
Sume's video generation docs list the request fields: prompt, duration, resolution, aspect_ratio, frame_images, input_references and generate_audio. There is no camera field, and provider.options is rejected for every model today, so camera intent has to live in prompt.
A workable pattern is one move per shot, stated as a verb phrase: a slow push in on the product, then a static hold. Keep a short test clip per move so you can see how the model reads it before spending on a 15 second render. Sume's Kling 3.0 Motion Control route is different again: it copies motion from a reference video rather than reading camera words.
A practical checklist
- Hailuo 2.3 on MiniMax's API: put
[Push in]or similar in the prompt, one command group per intent. - MiniMax H3 or Sume: write the move in words and verify on a 5 second clip first.
- Static shots: say static or locked-off explicitly, since there is no command to turn camera drift off.
- Combine at most two moves per clip until you have seen the model handle one.
Sources
Related posts
More in Models
- Hailuo first frame image limits: 20 MB, 300 px, 2:5 to 5:2
MiniMax's Hailuo image-to-video wants JPG, PNG or WebP under 20 MB with a short edge over 300 px; H3 frame images list 256-5760 px and 30 MB. Compare both.
- MiniMax video API error 1026 and 1027: sensitive content
MiniMax returns 1026 when the input is flagged and 1027 when the generated video is. What 1002, 1008 and 2013 mean, and how Sume reports failures.
- MiniMax video callback_url: echo the challenge in 3 seconds
MiniMax checks a callback_url by POSTing a challenge you must echo within 3 seconds. Sume skips the handshake and signs each terminal webhook instead.
- MiniMax video tasks are queryable for 7 days: poll every 10 s
MiniMax says video tasks can be queried for the last 7 days and suggests polling every 10 seconds. What that means for your pipeline, and Sume's 30 s example.
Written by Sume