MiniMax: Hailuo 2.3

MiniMax: Hailuo 2.3 is a video generation model from MiniMax, listed on api14 under the video category. It is built for turning written descriptions and reference stills into short video clips, and it sits in the Hailuo line of MiniMax's generative media models. If your product needs to produce motion footage from a prompt rather than edit existing video, this is the kind of model the catalog entry points to.

The model takes text and images as input and returns video as its only output. That means two main workflows are supported: text to video, where a prompt alone drives the result, and image to video, where a still is used as the starting point. Output is generated at 1080p in a 16:9 aspect ratio, and clips can be 6 or 10 seconds long. A first frame image can be supplied as the reference frame for image-to-video runs. Audio generation is not part of this model, so any sound you need has to be added downstream. No context length is listed for this entry.

Those capabilities map onto a fairly specific set of tasks. Marketing and social teams can generate short 16:9 spots from a script line or a product photo. Storyboard and previsualization work benefits from the image-to-video path, since a concept frame can be animated into a 6 or 10 second beat at 1080p. Game and app developers can produce looping or transitional footage without a shoot. Because clips are short and audio is not generated, the model suits shot-level production, not long-form assembly on its own.

You call it through api14's OpenAI-compatible API using the model id "minimax/hailuo-2.3". The supported request parameters are max_tokens, temperature and top_p, so you can shape sampling behaviour and output length in the same way as other models on the gateway. Because the schema follows the OpenAI format, existing client libraries and code paths generally work with only the model id, prompt and any reference image changed. Note that maxOutput is listed as 0 for this entry, and the fixed settings for resolution, aspect ratio, duration and first-frame input come from the model's own configuration rather than from additional parameters.

There is no context length recorded here, and no audio track is produced. If you need sound generated alongside the picture, a different model will fit better. If your work is long-form video or requires aspect ratios other than 16:9, or resolutions beyond 1080p, this model's listed options will not cover it. For short, silent, 1080p clips built from a prompt or a starting frame, it is a direct fit.