MiniMax requires a shape and never infers one
The reference lists 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, states that the field is required for text-to-video and that it is never adaptive. For image-to-video the ratio follows the supplied image, so the same row behaves in two ways. As of 2026-09-22.
| Path | Who decides the shape | What happens if the field is omitted |
|---|---|---|
| Text-to-video | The request, from six named ratios | The request is not valid; the field is required |
| Image-to-video | The supplied image | The input frame's shape is carried through |
| Either | Never the model | Adaptive is explicitly not offered |
Inclusion rule. Statements this reference makes about the aspect ratio field, separated by the input path each applies to. Ratios available on earlier model generations are not recorded. Order. Text path first, as the reference documents it.
1Required is a stronger statement than offered
Most rows in this column publish a list and leave the field optional with a default behind it. This one refuses to proceed without a choice, which removes a class of quiet error: nothing here can come back in an unexpected shape because nobody set the parameter.
It costs something in pipeline design. A batch job has to carry the shape per request rather than per campaign, and a variant set that runs across a vertical and a widescreen placement is two explicit sets rather than one set with a flag.
2Never adaptive, stated as a refusal
One other row in this register defaults to adaptive, letting the model choose. Here the page says adaptive is not available, so the two entries sit at opposite ends of the same question and a plan carried between them has to change shape handling entirely.
For advertising the explicit end is the safer one. An adaptive shape is a shape decided after the brief, and a placement that accepts one ratio does not care why the file arrived in another.
3Six shapes including both extremes
Twenty-one by nine and nine by sixteen on the same list covers the ultrawide and the fully vertical cases, with four by three, one by one and three by four between them. Only three rows in this register publish a square option and this is one of them.
The list applies to both models this vendor documents, which makes shape the one axis where the model choice does not matter. Resolution and duration both move between the two models; the six ratios do not.
4Where this cell sits in the column
| Generator | Aspect ratio |
|---|---|
| Amazon Nova Reel | Not stated |
| CogVideoX | Not stated |
| Google Veo | 16:9 and 9:16 |
| Grok Imagine | Seven, 16:9 the default |
| Hedra | Seven, including 9:21 and 21:9 |
| HeyGen | 16:9 or 9:16 |
| Kling AI | Not stated |
| LTX-Video | Not stated |
| MiniMax | Six, required for text-to-video |
| Pika | Not stated |
| PixVerse | Eight on version 6 |
| Runway | Seven pixel pairs |
| SceneMixer | 9:16 and 16:9 |
| Vidu | Not stated |
| Wan 3.0 | Six, adaptive by default |
Inclusion rule. Every generator in the register, carrying whatever its own page says about output shape. Six rows say nothing and are marked hollow rather than dropped. Order. Alphabetical by generator name.
- Aspect ratios; for image-to-video the ratio follows the input image21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, required and never adaptive for text-to-video
- Resolutions768P and 2K on MiniMax-H3; 480P and 768P on H3-Max, where 2K is not supportedstated per model
5Sources
Read from the MiniMax video task reference, checked 2026-09-22. The column itself is described on aspect ratio, and the rule that keeps surfaces and generators apart is on method. Next along this column: Wan 3.0, PixVerse.