AI Video Generation: From Prompt to MP4
Text-to-video and image-to-video through one async task API. Model comparison (MiniMax H3 vs Kling 3.0 Omni), polling code and cost control for per-second billing.
What you'll build
The same submit-and-poll pattern as images, pointed at video models: describe a scene or upload a still, and get back an MP4. This is the core of short-video generators, ad creative tools and storyboarding features.
Pick a model
| Model | Price | Strength |
|---|---|---|
| minimax-h3 | 12 Credits / second | omni-modal, generates native stereo audio |
| kling-3.0-omni | 21 Credits / second | text/image/reference-video modes in one family |
A 10-second clip costs 120 credits on MiniMax H3 or 210 on Kling 3.0 Omni. Failed tasks are not charged.
1. Submit
const headers = { Authorization: `Bearer ${process.env.ORUX_API_KEY}`, "Content-Type": "application/json" }
const submit = await fetch("https://orux.top/api/v1/tasks", {
method: "POST", headers,
body: JSON.stringify({
model: "minimax-h3",
input: { prompt: "a paper boat sailing down a rainy street gutter, ambient rain sound" },
}),
})
const { task_id } = await submit.json()2. Poll
let task
do {
await new Promise((r) => setTimeout(r, 10000))
task = await (await fetch(`https://orux.top/api/v1/tasks/${task_id}`, { headers })).json()
console.log(task.status)
} while (task.status === "QUEUED" || task.status === "RUNNING")
if (task.status === "SUCCESS") console.log("video:", task.result[0])Video renders take longer than images (typically tens of seconds to a few minutes depending on model and duration). Poll every 10 seconds and persist the task id — if your process restarts, you can resume polling instead of paying for a second submission.
Cost control
- Duration drives cost: a 15-second clip costs 2× the 7-second one.
- Decide the clip length before submitting; the model page lists each model's allowed durations.
- Show an estimated cost in your own UI before the user confirms — it prevents refund requests.
Next steps
Compare all video models on the model market, and see each model page for durations, resolutions and image-input options.