Introducing the Magi 2.0 API on Pixazo API: Image to Video and Video Editing

What Magi 2.0 actually is

Magi 2.0 is a high-performance video generation model built by Sand AI and served on Pixazo as the Magi 2.0 API. It turns a text prompt plus a reference into high-fidelity, cinematically detailed motion. Two things set it apart for real production work. First, spatial awareness: it renders physical interactions and fluid camera movement with unusual consistency, so objects hold together as they move. Second, control: you get direct handles on duration, resolution, frame rate, and sampling, instead of a single opaque button. On Pixazo the model exposes two operations, and you switch between them by changing the endpoint, not the key.

• Image to Video animates a single still into a clip that follows your prompt.
• Video to Video takes an existing clip and restyles it, with named artistic presets.

Image to Video, end to end

You send a prompt and an image_url, Magi animates the frame, and you poll a single status endpoint until the clip is ready. The whole path is asynchronous and built for servers, not for a browser tab waiting on a spinner.

Every parameter you can set

Image to Video takes seven fields. Only the prompt and the image are required; everything else has a sane default, so you can start simple and tighten later.

* required. A higher guidance_scale pushes the model to follow your prompt more literally. A fixed seed makes a result reproducible, which matters when you are iterating on a shot and want to change one thing at a time.

Video to Video: restyle an existing clip

The second operation takes a video_url instead of an image and rewrites it toward your prompt. It accepts longer clips (up to 30 seconds), a wider frame rate range (15 to 60 fps), and a style preset that steers the whole look in one word.

The async workflow

Both operations return immediately with a request_id and a status of QUEUED. You then poll one universal status endpoint, or register a webhook and skip polling entirely. A generation that ends in ERROR is a system error and is not charged.

Code: from image to finished clip

Authentication is a single header, Ocp-Apim-Subscription-Key. Here is the full round trip: submit, then poll until the status flips to COMPLETED.

# 1. Submit an image-to-video job
curl -X POST "https://gateway.pixazo.ai/magi/v1/v1/image-to-video" \
  -H "Content-Type: application/json" \
  -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY" \
  -d '{
    "prompt": "A paper boat drifting down a rain-soaked street, slow dolly-in, cinematic",
    "image_url": "https://example.com/boat.jpg",
    "duration": 6,
    "aspect_ratio": "16:9",
    "fps": 24,
    "guidance_scale": 7.5
  }'

# Response: { "request_id": "magi_01a01a89-615d-746c-ae38-37d9f40759e2c", "status": "QUEUED", "polling_url": "..." }

# 2. Poll until COMPLETED
curl -H "Ocp-Apim-Subscription-Key: YOUR_API_KEY" \
  "https://gateway.pixazo.ai/v2/requests/status/magi_01a01a89-615d-746c-ae38-37d9f40759e2c"

# COMPLETED in ~2m40s: output.media_url -> ["https://pub-582b...r2.dev/v1/magi_01a01a89-.../output.mp4"]

Pricing is a flat per-second rate

There is no token math to reason about. Magi bills by output length at a single 720p rate, so cost scales linearly with how long your clip is.

Where a video generation API earns its place

Why run Magi on Pixazo

Magi does not live alone on Pixazo. It sits behind the same gateway, the same key, and the same async contract as every other model, so adding it to an app that already calls another video model is a change of endpoint, not an integration project. The universal status endpoint, the webhook option, and the standard error codes are identical across the catalog.

That makes it easy to route by job: reach for Magi when a shot needs physical plausibility and camera motion, and swap to a sibling when you need something else. A few neighbours worth knowing:

Wan → Text and image driven generation.Kling → Fast, expressive short-form motion.Seedance → Stylised, character-led clips.

If you would rather try prompts by hand before writing a line of code, the AI image-to-video generator runs these models from one prompt box.

Field notes for good results

• Feed a clean, well-lit input image. Magi animates what is there, so a sharp subject on an uncluttered background moves more convincingly than a busy frame.
• Write motion into the prompt. Name the camera move (dolly, pan, orbit) and the action, not just the scene, since spatial awareness is where this model earns its keep.
• Start at the default duration of 5 seconds while you dial in the look, then extend once the shot is right. You pay by the second.
• Lock the seed when iterating so you can change one variable at a time and actually see its effect.
• For Video to Video, let the style preset do the heavy lifting and keep the prompt focused on the subject rather than the aesthetic.

Get started

Create a key, send one image, and read back a clip. That is the entire first run.

1. Sign up and create a free key at the Pixazo API console.
2. POST a prompt and an image_url to https://gateway.pixazo.ai/magi/v1/v1/image-to-video with your key in the Ocp-Apim-Subscription-Key header.
3. Poll the returned polling_url until the status reads COMPLETED, then download output.media_url.

The takeaway

Magi 2.0 brings Sand AI’s physics-aware motion to Pixazo with two clean operations, honest per-second pricing, and the same production plumbing as the rest of the catalog. Point it at a still to bring a moment to life, or point it at a clip to change its whole look. Either way it is one key away.