MiniMax Hailuo 2.3 Tutorial: AI Video Creation

MiniMax Hailuo 2.3 is a powerful tool for creating AI-generated videos with realistic motion and cinematic effects. It is available on APIMart through the MiniMax Hailuo 2.3 API, with two workflows: text-to-video and image-to-video from a single first-frame image. That makes it suitable for developers, studios, and educators. Here's what you need to know:

• Key Features: Generate videos in 768p or 1080p resolution, with durations of 6 or 10 seconds at 768p and 6 seconds at 1080p. MiniMax says 2.3 improves body movement and micro-expressions over earlier versions.
• Pricing: As of October 2026, prices start at $0.0248 per second for Fast at 768p. Fast costs about half of the standard model at 768p and about 40% less at 1080p.
• Setup: Sign up on APIMart, generate an API key, and use a simple three-step process: submit a task, poll for status, and download the video from the task result.
• Optimization Tips: Use the Fast model for image-to-video drafts, switch to Standard when you need text-to-video or final renders, and write clear prompts using the CCR (Camera, Character, Reaction) framework.

This guide simplifies the video creation process, ensuring quality results while managing costs efficiently.

Key Takeaways

• MiniMax Hailuo 2.3 on APIMart generates 768p or 1080p clips through two workflows: text-to-video and image-to-video from a single first-frame image. There is no last-frame or subject-reference input.
• Every workflow follows the same asynchronous three-step process: submit a POST request to get a task_id, poll GET /v1/tasks/{task_id} every 10 seconds until the status is completed, then download the MP4 from data.result.videos[0].url[0]; results are kept for 72 hours.
• As of October 2026, APIMart prices Hailuo 2.3 from $0.0248/sec for MiniMax-Hailuo-2.3-Fast at 768p to $0.072/sec for the standard model at 1080p.
• The prompt_optimizer parameter defaults to true; set it to false for exact wording, and use the 15 bracketed Chinese camera commands such as [推进] (push in) at the start of the prompt.
• 768p supports 6 or 10 seconds, 1080p is limited to 6 seconds, prompts are capped at 2,000 characters, and the Fast model requires first_frame_image while still requiring a prompt.
• Fast costs about half of Standard at 768p and about 40% less at 1080p, so prototype image-to-video in Fast at 768p and switch to the settings you need for final renders.

Setting Up MiniMax Hailuo 2.3 on APIMart

Creating and Configuring an APIMart Account

To get started, head over to apimart.ai and sign up for a free account. Once logged in, navigate to the "API Key Management" section on your dashboard. Generate a new API key, and make sure to copy and save it immediately since it will only be displayed once^[5].

Next, search for MiniMax-Hailuo-2.3 or MiniMax-Hailuo-2.3-Fast in the model dashboard or API documentation. This lets you confirm its availability and review the endpoint details before crafting your first API request.

Once your account is ready and your API key secured, the next step is setting up your development environment.

Development Prerequisites

APIMart's API supports multiple languages, including Python (via requests), JavaScript/TypeScript (via axios), and cURL. Requests are formatted in JSON and require a Bearer token for authentication.

To keep your API key safe, store it in an environment variable like os.environ["APIMART_API_KEY"] instead of hardcoding it into your scripts.

Video generation is an asynchronous process. Here's how it works: you submit a task to POST https://api.apimart.ai/v1/videos/generations, receive a task_id in data[0].task_id, poll GET https://api.apimart.ai/v1/tasks/{task_id}, and then download the video from data.result.videos[0].url[0] once the status is completed. APIMart does not publish a fixed render time; it depends on resolution, duration, and current load^[3].

With your environment set up and a clear understanding of the workflow, you can focus on managing your budget and optimizing usage.

Tracking Budget and Setting Usage Limits

Here are APIMart's per-second prices for each variant as of October 2026^[6]; check the model page for current rates.

For example, generating a 6-second 768p clip with the standard model costs around $0.29, while using the Fast variant reduces it to about $0.15, roughly half. At 1080p, a 6-second clip costs about $0.43 on Standard and $0.25 on Fast, about 40% less. A practical approach is to prototype image-to-video with MiniMax-Hailuo-2.3-Fast at 768p and then move to the resolution and model you need for final renders. Check the model page for current prices.

APIMart's Billing dashboard allows you to monitor your spending in real time, which is particularly handy when running batch jobs since costs are calculated per second of video generated.

Core Video Generation Workflows

Text-to-Video Workflow

Creating videos from text follows a straightforward three-step process: submit, poll, and download.

• Create the task: Send a POST request to https://api.apimart.ai/v1/videos/generations with model, prompt, duration, and resolution. Text-to-video uses the standard MiniMax-Hailuo-2.3 model, because Fast requires a first-frame image. In return, you'll receive a task_id in data[0].task_id.
• Poll for status: Use the task_id to query GET /v1/tasks/{task_id} every 10 seconds. While the task is in progress, the status will be pending or processing. Once it finishes, the status changes to completed.
• Download the video: Read the video URL from data.result.videos[0].url[0] and save the MP4 to your own storage. Results are kept for 72 hours.

One important feature here is the prompt_optimizer parameter. By default, this is set to true, meaning the model will refine your prompt to improve the visual quality of the video. However, if you need exact control - for example, in branded content where precise wording is critical - you can set it to false ^[3].

For camera movements, Hailuo 2.3 on APIMart supports 15 bracketed Chinese commands placed at the start of the prompt with the exact labels: [左移] / [右移] (pan left / right), [左摇] / [右摇] (rotate left / right), [推进] / [拉远] (push in / pull out), [上升] / [下降] (rise / descend), [上摇] / [下摇] (tilt up / down), [变焦推近] / [变焦拉远] (zoom in / out), plus [晃动] (shake), [跟随] (follow), and [固定] (static) ^[3].

Building on this text-based approach, the image-to-video workflow offers even more control by anchoring your video to a specific starting image.

Image-to-Video Workflow

The image-to-video process uses the first_frame_image parameter, which accepts either a public URL or a Base64 data URI. It works with both MiniMax-Hailuo-2.3 and MiniMax-Hailuo-2.3-Fast; for Fast, first_frame_image is required and the prompt is still required. MiniMax's own docs recommend JPG, JPEG, PNG, or WebP files under 20MB, with a short edge of at least 300px and an aspect ratio between 2:5 and 5:2 ^[2].

This workflow follows the same three-step structure as text-to-video. The difference is that your provided image sets the initial frame, while your text prompt dictates how the scene evolves. This makes it ideal for scenarios like marketing or education, where you might want a product image or diagram to transition into an animated sequence.

A helpful trick for creating longer videos is to take a screenshot of the last frame of a completed clip and use it as the first_frame_image for the next task. This helps keep character and scene consistency across multiple clips without requiring additional adjustments ^[9].

What Hailuo 2.3 Does Not Support on APIMart

Hailuo 2.3 on APIMart takes only one image input, first_frame_image. It has no last_frame_image and no subject-reference input, so first-and-last-frame transitions and face-reference workflows are not available on this model ^[3].

If you need those, look at other MiniMax models on APIMart: MiniMax Hailuo 02 API supports first-and-last-frame transitions, and MiniMax H3 API supports first/last-frame input plus multimodal reference images, video, and audio.

Both Hailuo 2.3 workflows - text-to-video and image-to-video - share the same three-step asynchronous process and camera command syntax. Switching between them is as simple as adding or removing first_frame_image in your POST request.

Hailuo AI Video Masterclass: From Beginner to Pro (Full Guide 2.3)

Improving Video Quality and Cutting Costs

Writing Effective Prompts

Crafting a clear and precise prompt is essential for producing high-quality video output. A useful method to structure your prompts is the Camera, Character, Reaction (CCR) framework. This breaks down the scene into three components: what the camera is doing, who is in the shot, and what action is taking place. For example: "[跟随] Camera: slow follow shot; Character: a quarterback in a blue jersey; Reaction: throwing a deep pass during a snowy night game in Chicago, stadium lights creating a hazy glow."

Adding specific details about the visual style and focus can make a big difference. Terms like "photorealistic", "cinematic lighting", or "anime style" guide the model toward your desired look. Including subtle character details, such as "a slight eyebrow raise" or "a thoughtful gaze", lets you use the micro-expression improvements MiniMax highlights for Hailuo 2.3 ^[4]. However, avoid cramming too many actions into a single prompt, as this can lead to awkward or glitchy motion. With a 2,000-character limit, aim for prompts that are detailed yet streamlined for clarity and purpose ^[3][7].

Choosing the Right Duration and Resolution

When deciding on video resolution and duration, it's important to weigh your options. Hailuo 2.3 offers two resolutions: 768p and 1080p. The key difference? 1080p clips are limited to 6 seconds, while 768p supports both 6-second and 10-second durations ^[3].

For initial drafts, 6-second clips at 768p are a practical choice. They're affordable, allowing you to evaluate motion and composition without overcommitting resources. Once you've narrowed down your options, you can switch to higher-resolution settings for the final product.

Using an Iterative Workflow

An iterative workflow is key to balancing quality and cost. For image-to-video, a two-stage process works well: explore with the Fast model, then re-run the best version with the settings you need for delivery.

"Generate 3–5 variations of the same prompt using Hailuo 2.3 Fast during concept phase. Pick your best, then re-run that exact prompt in 2.3 Standard or 02 at 1080p for final output. You'll burn fewer credits on failed experiments." - QWE AI Academy ^[8]

On APIMart, a 6-second 768p clip costs about $0.15 on Fast versus about $0.29 on Standard, so Fast is roughly half the price at 768p and about 40% cheaper at 1080p. MiniMax describes Fast as faster and more cost-effective for batch creation ^[4]. By testing multiple drafts in Fast, you can identify the most promising version before spending more on the final render. Compare the outputs on your own prompts before choosing a tier for delivery.

Integrating MiniMax Hailuo 2.3 Videos into Production

Managing Asynchronous Tasks and Output Files

Using Hailuo 2.3 for video generation involves a three-step asynchronous process: first, submit a request and receive a task_id. Next, poll GET /v1/tasks/{task_id} until the status is completed. Finally, read the video URL from data.result.videos[0].url[0] and download it.

If you're polling, stick to a 10-second interval to avoid hitting rate limits. For larger batches, keep pending task IDs in a queue and poll them on a shared schedule instead of starting a tight loop for each job.

Keep an eye on the status field for potential errors. If it shows failed, read the error field immediately for troubleshooting or logging. Download your files promptly; results are kept for 72 hours. To keep track of tasks across asynchronous responses, store each task_id alongside your internal production ID so you can map results back to requests.

By setting up an efficient task management system, you'll have a smoother experience controlling costs and scaling operations.

Budgeting and Scaling Your Usage

Hailuo 2.3 is billed per second of generated video, so cost calculations are straightforward. For instance, 2,400 seconds of Fast output at 768p (as of October 2026, $0.0248/sec) would cost about $59.52, while the same volume on the standard model at 768p would cost about $117.12.

To keep costs manageable, consider using Fast for initial image-to-video renders. As of October 2026 on APIMart, Fast is about half the price of Standard at 768p and about 40% cheaper at 1080p. The task query response also includes a cost field, so you can log the real cost of each task without waiting for monthly invoices.

"Hailuo 2.3 Fast model... generates videos faster at a lower price, reducing costs for batch creation by up to 50%." - MiniMax News ^[4]

When scaling production, choose the polling approach that matches your workload:

If you're handling many concurrent tasks, a queue-based worker is the better option. A simple polling loop works fine for small-scale or one-off jobs.

Conclusion

MiniMax Hailuo 2.3, available on APIMart from $0.0248 per second (Fast at 768p, as of October 2026), brings professional AI video production within reach for a fraction of the cost. For example, a 6-second 768p clip costs about $0.15 on Fast and about $0.29 on Standard.

To get started, set up your APIMart account and choose the workflow that matches your creative goals: text-to-video on Standard, or image-to-video on Standard or Fast. Fine-tune your results using the CCR (Camera, Character, Reaction) method^[11]. For longer projects, keep visual consistency by capturing the final frame of each clip and using it as the starting frame for the next segment^[9].

The Fast model offers clear savings: about half the price of Standard at 768p and about 40% less at 1080p. Use it for drafts and batch runs, and pick the model and resolution for final renders after comparing the outputs on your own prompts^[4].

This cost efficiency hasn't gone unnoticed:

"Hailuo 2.3 once again sets a new global record for video model cost-effectiveness... offering 'more for the same price' to both business and consumer users." - MiniMax Official News^[4]

For teams managing larger-scale video production, the integration strategies outlined above are invaluable. Queue-based polling, the per-task cost field, and prompt templates make it easier to run a scalable production pipeline for multiple clips each week.

FAQs

What's the best way to keep characters consistent across multiple clips?

Hailuo 2.3 on APIMart has no subject-reference input, so consistency comes from your first frames and prompts. Start each clip from a first_frame_image with uniform lighting and angles, and reuse the last frame of one clip as the first frame of the next. Be precise in your prompts, detailing the subject, action, and style clearly. Begin with short, six-second clips to test consistency before moving on to longer or higher-resolution videos.

How do I check the status of many video jobs at once?

For a few jobs, poll each one with GET /v1/tasks/{task_id} every 10 seconds or so.

For larger batches, store the pending IDs in a queue and let one worker poll them on a shared schedule, backing off when you see HTTP 429. That keeps the number of requests predictable and keeps you under rate limits.

Why would I turn off prompt_optimizer?

When you want precise control over video generation, set the prompt_optimizer parameter to false. This stops the system from automatically tweaking your prompts, ensuring your exact wording and specifications directly guide the model's output.