How to Prompt Grok Imagine Video 1.5

XONA's latest video endpoint is live: POST /video/grok-1.5-short, powered by xAI's grok-imagine-video-1.5. It takes an image and a prompt, and generates a high-quality video clip — complete with synchronized audio — in a single pass. Any agent with a Solana wallet can call it for $0.50 via x402, no subscription, no API key, no billing account required.

But like any frontier model, the quality of what you get is directly shaped by how you ask. This is a practical guide to getting the best results.

Start with a Still Image

The model is image-to-video. Your starting frame is the foundation — composition, lighting, and subject placement are all set before the model begins. Get those right first.

A useful workflow: generate a still image using /image/grok-imagine or any image generation endpoint, then pass that URL directly into /video/grok-1.5-short. Your video prompt only needs to describe what changes — what moves, what sounds, what happens next.

This separation of concerns keeps your prompts clean and your results predictable.

Design the Sound Like a Professional

grok-imagine-video-1.5 generates audio alongside video in a single pass. That means your prompt influences what you hear, not just what you see.

Vague sound descriptions produce vague results:

"Sound: city sounds, traffic"

Specific spatial cues produce specific results:

"Sound: cars passing by, skateboards rolling on pavement, teenagers laughing, the distant rumble of a delivery truck"

Think about foreground versus background. Think about texture, distance, and layering. The model responds to the same specificity a sound designer would expect in a brief.

Use Intensity Modifiers

How something moves matters as much as what moves. The model distinguishes between ambiguous motion and described motion progression.

Weak:

"the wave crests"

Strong:

"the wave crests fully and pitches forward, the translucent green crest folding and crashing down with tremendous force, white foam churning across the entire frame"

Apply the same logic to character motion, environmental shifts, and transitions. Words like slowly, then faster, with force, barely perceptible, and surging all carry real weight.

Specify Camera Movement

Left unspecified, the model defaults toward static or minimal camera movement. If you want something cinematic, ask for it explicitly.

Effective camera directions:

Slow push-in — draws the viewer toward the subject over time
Aerial approach — wide establishing shot descending toward a subject
Tracking shot — follows a subject laterally as it moves through frame
Locked static frame — no camera movement; often reads as the most cinematic choice for intimate scenes
Gentle drift — subtle environmental motion without a hard direction

If your scene calls for stillness, state it. If it calls for movement, name the technique.

Keep Prompts Focused

The model handles concentrated narratives better than sprawling ones. One clear scene beat with defined roles — who is doing what, where, in what lighting — outperforms a prompt that tries to describe too much at once.

A focused prompt:

"A woman sits at a café table near a rain-streaked window, holding a ceramic cup with both hands. She looks slowly out toward the street. Sound: light rain on glass, ambient café murmur, distant traffic."

A scattered prompt:

"A woman is in a café and it's raining and she's thinking about something and there are people around and maybe a car goes by and there might be music."

Same scene. Very different results.

Calling the Endpoint

Any x402-compatible agent can call this endpoint directly:

POST https://api.xona-agent.com/video/grok-1.5-short

Required:

prompt — your scene description, including a Sound section
image_url — your starting frame (.jpeg, .jpg, .png, or .webp)

Optional:

resolution — 480p, 720p, or 1080p (default: 720p)
aspect_ratio — e.g. 16:9, 9:16, or auto (default: auto)

The endpoint is discoverable via xpay.discover("video generation") using the xPay SDK, and payment settles automatically on Solana via x402 before the generation runs.

What You Get Back

{
  "video_url": "https://cdn.xona-agent.com/generated/grok-imagine15-....mp4",
  "thumbnail_url": "https://cdn.xona-agent.com/generated/grok-imagine15-....jpg",
  "duration": 10,
  "model": "grok-imagine-video-1.5",
  "metadata": {
    "prompt": "...",
    "image_url": "...",
    "aspect_ratio": "auto",
    "resolution": "720p"
  }
}

A CDN-hosted .mp4, a thumbnail, and the metadata. No intermediate URLs, no polling required — the response comes back when the video is ready and uploaded.

The best video prompts read like a film brief: they name the subject, describe the motion, specify the sound, and call out the camera. Start there, and the model has everything it needs.

Try it: xona-agent.com/resources