Hailuo Video API: Lifelike Motion Video by MiniMax

Hailuo API 通过一个兼容 OpenAI 的密钥，在 Atlas Cloud 上为开发者提供 MiniMax 的视频模型。它涵盖了 Hailuo 02 和 Hailuo 2.3 系列的文生视频和图生视频功能，核心优势在于逼真且符合物理规律的运动以及强大的提示词控制。如果您需要逼真的动态效果或专业的动漫和插画风格，它将是理想的选择。

Hailuo 02 专注于逼真的电影级输出，具备顶尖的物理仿真效果和极高的指令遵循能力。Hailuo 2.3 专攻动漫和插画视频，严格遵循动画原理而非仅仅进行近似模拟。对于照片级逼真的工作请使用 02，而当需要呈现真正动画质感的视觉效果时请使用 2.3。

这两条产品线都支持文本提示词，图生视频功能接受参考图像以锚定构图和氛围。输出的视频分辨率最高可达原生1080p，时长大约为5到10秒。Hailuo仅生成视觉内容，因此您需要在编辑时单独添加音乐、语音或特效。

Hailuo 可生成最高原生 1080p 的视频片段，其中 Standard（标准）版本为 768p，Pro（专业）版本为全 1080p。根据模型和版本的不同，时长在 5 到 10 秒之间。该模型不支持 4K 输出，因此请将需要更高分辨率的工作交由专为此构建的模型处理。

不。Hailuo 纯粹专注于视觉效果，不生成原生音频。对于完成的片段，请在后期制作中将输出结果与您自己的音乐、画外音或音效进行合成，或者使用单独的音频模型。如果您需要同时生成视频和音频，具备原生音频功能的模型会是更好的选择。

Standard balances quality and cost for everyday clips, Pro pushes native 1080p detail for final and commercial work, and Fast trades a little fidelity for quicker, cheaper runs. All tiers share the same prompt and image inputs, so you can move a job between them without reworking your request.

Hailuo 02 reports a high response rate on complex, multi-part instructions, so cameras, actions, and scene details tend to land as written. Repeated calls with the same prompt also stay reasonably consistent, which helps client work and pipelines that need comparable output across runs.

Yes, that is Hailuo 2.3's specialty. It commits to animation conventions, exaggerated arcs, anticipation, held poses, and anime-standard color, instead of the semi-realistic compromise general models produce. That makes its output easier to blend with hand-drawn or traditionally animated material.

Generation is asynchronous: each request returns a prediction ID that you poll until the clip is ready, which fits queues and high-volume runs. Use the Fast tiers for iteration and previews, reserve Pro for finals, and add retry logic with backoff so one failed job does not stall the batch.

Uploads that contain real human faces are subject to platform content rules and identity protections, and may be restricted. For a consistent look, build characters from prompts or non-identifying reference images rather than a real person's photo, and review Atlas Cloud's acceptable use terms before relying on face-based inputs.