Image to Video API: How to Animate Images Programmatically (2026 Guide)

An image to video API takes a still image and returns a short clip that starts from it. Compared with text to video, you give up some freedom (the first frame is fixed) and gain a lot of control: the subject, composition, colors and branding are exactly what you supplied. That makes it the usual choice for product shots, character animation and anything where the look has to match an existing asset.

This guide covers the parts that decide whether the output is usable: the source image, the motion prompt, the model, and the code to run it.

Most bad image-to-video results trace back to the input, not the model.

• Match the aspect ratio. If you want a 16:9 clip, send a 16:9 image. When the shapes differ, the model has to crop or invent the missing area, and the first frame no longer matches your image exactly. Pad (letterbox) the image yourself when the edges must stay intact, as with product photos; crop it when they do not matter.
• Use a clean, well-lit subject. Motion amplifies problems. Blurry edges, heavy compression and cluttered backgrounds turn into flicker once animated.
• Make it publicly reachable. The API downloads the image from the URL you send, so it must be accessible without authentication. JPG, PNG, WebP and GIF are accepted. A pre-signed URL from your storage bucket works well.

The image already defines the scene, so the prompt should describe what moves and how:

• Name the motion of the subject: "she turns her head toward the window and smiles", "steam rises from the cup".
• Name the camera move separately, using film vocabulary: "slow push in", "static camera", "gentle pan left", "orbit around the product".
• Keep it modest. A fast camera move combined with fast subject motion is the most common cause of warped faces and stretched backgrounds. Pick one to emphasize.
• Do not describe things that would contradict the image (a different outfit, a new background). The model has to reconcile the conflict, and usually does so badly.

The prompt is optional on AirCube, but leaving it out hands every motion decision to the model.

import os
import time

import requests

API_KEY = os.environ["AIRCUBE_API_KEY"]
BASE = "https://aircube.ai/api/v3"
HEADERS = {
    "Authorization": f"Bearer {API_KEY}",
    "Content-Type": "application/json",
}


def animate(model: str, image_url: str, prompt: str, duration: str = "5", resolution: str = "720p") -> str:
    response = requests.post(
        f"{BASE}/{model}/image-to-video",
        headers=HEADERS,
        json={"image": image_url, "prompt": prompt, "duration": duration, "resolution": resolution},
        timeout=30,
    )
    response.raise_for_status()
    generation_id = response.json()["data"]["id"]

    deadline = time.time() + 600
    while time.time() < deadline:
        status = requests.get(f"{BASE}/status/{generation_id}", headers=HEADERS, timeout=30)
        status.raise_for_status()
        data = status.json()["data"]
        if data["status"] == "completed":
            return data["output_url"]
        if data["status"] == "failed":
            raise RuntimeError(f"Generation {generation_id} failed (the charge is refunded)")
        time.sleep(5)
    raise TimeoutError(generation_id)


clip_url = animate(
    "kling-v3.0-pro",
    "https://example.com/products/sneaker-side.jpg",
    "The sneaker rotates slowly on a turntable, static camera, soft studio light",
)

Rendering takes from about 30 seconds to a few minutes, so in a web app return the generation id to the client immediately and do the polling in a background worker. The returned output_url is a signed link valid for 7 days; copy the file to your own storage if you need to keep it.

Prices are for a 5-second clip at 720p as of October 2026; see pricing for other durations, resolutions and the current numbers.

Some rules of thumb:

• Test motion prompts cheaply. Hailuo 02, HappyHorse 1.0 or Seedance 2.0 Mini at 480p or 720p cost well under a dollar per attempt. Lock the prompt there, then render the final version on a stronger model.
• Longer clips from one image. Seedance 2.5 and Wan 3.0 go up to 30 seconds. For other models, see the next section.
• Several reference images. If you need the same character or product across shots, use reference-to-video instead, which accepts up to 9 reference images (supported by the Seedance models, Wan 3.0, Wan 2.6 and Veo 3.1). See the reference media guide.

Two options when one clip is not long enough:

1. Extend the clip. The Seedance models and Veo 3.1 support a video-extend task, which takes an existing video and continues it.
2. Chain from the last frame. Extract the final frame with FFmpeg (ffmpeg -sseof -0.1 -i clip.mp4 -frames:v 1 last.jpg), upload it, and use it as the source image for the next clip. The cut is less seamless than an extension but works with any model.