Runway Image to Video AI: Pricing, API and Code (2026)

Runway image to video AI takes a single still picture, such as a product shot, a portrait or a landscape, and turns it into a short clip in which the scene moves while keeping the look of the original frame. You supply a start frame and a sentence describing the motion, and the model generates everything that happens after that first frame. Below: how the feature works, how to get good results in the browser, how to call it from your own code, and how it compares on cost with Kling, Grok Imagine and PixVerse.

What is Runway image to video AI?

Runway is a company that builds generative video models and a creative web app around them. Its tools cover text-to-video, image-to-video and video-to-video editing. In the app you open a generative session, upload an image, describe your shot and generate. Runway image to video AI is the name people search for that middle option: start from a picture you already have, end with a moving clip.

Two terms are worth pinning down:

• Image-to-video: the model receives your image as the first frame, often called the start frame, and generates the frames that follow. The composition, colours and subject come from you.
• Text-to-video: the model builds the whole clip from a written description. It decides what the scene looks like, which gives you less control.

That control is the reason image-to-video is popular for real work. A brand team can keep the exact product photo that legal already approved and add a slow camera move. An illustrator can animate a finished piece without the model redesigning the character. A marketer can turn a catalogue shot into a short social clip with cinematic lighting and movement.

You do not have to use Runway’s own app to get this. On Unicorn API, Runway is available as the model id runway, with input fields for prompt, image_url, quality, duration, aspect_ratio and watermark. The same account also gives you other image-to-video models, which matters once you start comparing results and cost.

How does image-to-video AI actually work?

Almost every modern video generator, Runway included, is built on a diffusion model. A diffusion model learns to reverse noise: during training it sees real video gradually buried in random static, and it learns to remove that static step by step. At generation time it starts from pure noise and denoises toward a clip that matches its guidance. Most systems do this in a compressed representation of the video rather than on raw pixels, an approach called latent diffusion, which keeps the computation manageable.

In image-to-video, the guidance has two parts:

1. The start frame, which is a hard constraint. Whatever is visible in it should stay consistent: the face, the logo, the lighting direction.
2. The prompt, which is a soft constraint. It tells the model what should move and how, and the model interprets it.

This split explains most of what you will see in practice. Anything visible in your image is anchored. Anything not visible, such as the back of a product, a hand that was cropped out, or the room behind a closed door, has to be invented. Invented content is where artifacts appear: extra fingers, warped text, objects that melt as the camera turns.

Clip length matters for the same reason. Each frame is further from the anchor than the one before, so longer clips drift more. The output is played back at a fixed frame rate (frames per second), and more seconds means more frames the model has to keep consistent. Short clips with a single clear motion hold together best.

How to use Runway image to video AI in the browser

If you want to try the feature before writing any code, the browser is the quickest route. The steps are the same whether you use Runway’s app or the Unicorn API video playground, which lets you run Runway and other models against the same image with one account.

1. Pick a strong start frame. Use a sharp, well-lit image with one clear subject. Crop it to the aspect ratio you want the video to have, so the model does not have to pad or stretch it.
2. Upload it. In Runway’s app you tap to upload the image into a generative session. In the playground you choose the model and add the image URL or file.
3. Describe the shot. Write what moves, not what the image contains. The model can already see the image. More on this in the next section.
4. Set duration and aspect ratio. Start short. A short test tells you whether the motion works before you pay for a longer clip.
5. Generate, review, adjust one thing. If the result is off, change a single variable (the camera move, the speed, the crop) and run again. Changing everything at once makes it impossible to learn what helped.

Treat the first few runs as calibration. Every model reacts to prompts a little differently, and ten minutes of deliberate testing on one image teaches you more than an hour of random attempts.

How to write motion prompts for image to video

The most common reason a Runway image to video AI clip looks wrong is a prompt that describes the picture instead of the movement. “A red sports car on a mountain road at sunset” adds nothing; the model already has that. “Slow tracking shot following the car as it rounds the bend, dust lifting behind the rear wheels” gives it something to do.

A reliable structure has four parts:

• Camera: how the viewpoint moves.
• Subject: what the main subject does.
• Environment: what the surroundings do (wind, water, light, crowds).
• Pace: slow, gentle, sudden, real-time, slow motion.

Camera terms worth knowing

• Push-in / dolly in: the camera moves physically toward the subject. Good for product reveals.
• Pull-out: the reverse, revealing more of the scene. Risky, because the model must invent what was outside the frame.
• Pan: the camera rotates left or right from a fixed spot.
• Tilt: the camera rotates up or down from a fixed spot.
• Orbit: the camera circles the subject. Impressive, but it exposes sides the image never showed.
• Static camera: no camera motion; only the subject or environment moves. The safest option for faces and text.

Weak versus strong prompts

Start frame	Weak prompt	Stronger prompt
Coffee mug on a table	A beautiful cup of coffee	Static camera, steam rising slowly from the mug, soft morning light shifting across the table
Portrait of a woman	Woman smiling, cinematic	Slow push-in, she turns her head slightly toward the camera and smiles, hair moving in a light breeze
Rocky coastline	Epic ocean landscape	Slow drone shot gliding forward over the rocks at sunset, waves breaking in the foreground
Sneaker on a plinth	Cool shoe ad	Gentle orbit of about a quarter turn around the shoe, studio lighting, dark background

Keep one main motion per clip. “Push in while orbiting while the subject walks toward camera” asks the model to resolve three conflicting movements, and it usually resolves them badly. If you need a complex sequence, generate several short clips and cut them together in an editor.

Calling Runway image to video AI from code

Once you know which prompts work, the next step for many teams is automation: animating a whole product catalogue, generating variants for ads, or adding a “make it move” button to an app. Unicorn API runs Runway image to video AI as an asynchronous job. You submit a request, receive a job id, then poll until the clip is ready.

First, create an API key and export it as UNICORN_API_KEY. Then submit a job. This curl call uses the documented request shape for Runway, with your image passed as image_url:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "runway",
    "input": {
      "prompt": "A slow drone shot over a rocky coastline at sunset, waves breaking",
      "image_url": "https://example.com/coastline.jpg",
      "quality": "example",
      "duration": 1
    }
  }'

The quality and duration values above are the placeholders from the minimal request. Replace them with the values listed on the Runway model page and in the API documentation, where you will also find the allowed aspect_ratio options and how watermark behaves.

Here is a complete Python script that submits the job, polls it and prints the video URL:

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {"Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}"}

# Use values listed on /models/runway for quality and duration.
QUALITY = "example"
DURATION = 1

body = {
    "model": "runway",
    "input": {
        "prompt": "Slow push-in, steam rising from the mug, soft morning light",
        "image_url": "https://example.com/mug.jpg",
        "quality": QUALITY,
        "duration": DURATION,
    },
}

resp = requests.post(f"{API}/v1/jobs", json=body, headers=HEADERS, timeout=30)
resp.raise_for_status()
job_id = resp.json()["id"]
print("Submitted job", job_id)

deadline = time.time() + 15 * 60  # give up after 15 minutes
while time.time() < deadline:
    job = requests.get(f"{API}/v1/jobs/{job_id}", headers=HEADERS, timeout=30).json()
    status = job["status"]
    if status == "succeeded":
        print("Video ready:", job["outputs"][0]["url"])
        break
    if status == "failed":  # see /docs for the full list of statuses
        raise RuntimeError(f"Job failed: {job}")
    time.sleep(5)
else:
    raise TimeoutError("Job did not finish in time")

A few practical notes:

• Check the price first. Every request can be priced before it runs with POST /v1/models/price. Use it in batch jobs so a misconfigured duration or resolution never surprises you.
• Download the output. Save the file from job.outputs[0].url to your own storage rather than hot-linking it.
• Switching models is one field. To try Grok Imagine Video 1.5 on the same image, change the model id and use its field names: {"model": "grok-imagine-video-1-5-preview", "input": {"prompt": "...", "image": "https://example.com/mug.jpg", "duration": 5, "resolution": "720p", "aspect_ratio": "16:9"}}.
• Agents can do this for you. If you work in Cursor or Claude Code, you can connect the API through MCP and ask the agent to generate and compare clips.

Runway vs Kling, Grok Imagine and PixVerse: cost and fit

Runway is not the only strong image-to-video model, and the right choice depends on clip length, resolution and how much control you need over the ending. On Unicorn API, credits from monthly plans cost $0.01 each (Starter $25, Creator $50, Studio $100, Pro $200 per month), and top-up credits cost $0.015 each and never expire. Prices are rounded up to whole credits.

Model	How it is priced	Example clip	Credits
Runway	Flat per video	One video	12
Grok Imagine Video 1.5	Per second, by resolution	5 s at 720p	23 (22.5 rounded up)
Grok Imagine Video 1.0	Per second, by resolution, 6 to 30 s	6 s at 720p	27
PixVerse V6 Max Image To Video	Per second, by quality and audio	5 s at 720p, no audio	44
Kling 4.0 Keyframe To Video	Per second, by resolution	5 s at 720p	143 (142.5 rounded up)

When each one makes sense

• Runway at 12 credits per video works out to $0.12 on plan credits or $0.18 on top-ups. A Starter plan’s 2,500 credits covers about 208 Runway clips, which makes it a cheap default for animating catalogue images in bulk.
• Grok Imagine Video 1.5 is image-to-video only and very cheap per second, with strong motion quality. It runs 1 to 15 seconds at 480p, 720p or 1080p, and supports 1:1, 16:9, 9:16, 3:2 and 2:3. A 5-second 480p clip is 12 credits, the same as a Runway video. It can be busy at peak times, so build retries into batch jobs.
• Grok Imagine Video 1.0 covers longer clips, from 6 to 30 seconds, at the same per-second rate, and also does text-to-video.
• PixVerse V6 Max Image To Video can generate audio with the clip. At 720p it costs 8.8 credits per second without audio and 12 with it, and it takes a seed so you can reproduce a result.
• Kling 4.0 Keyframe To Video is the most expensive here, at 28.5 credits per second at 720p and 42.5 at 1080p, but it is the only one that takes a first frame, a last frame and extra keyframes. If people searching “kling ai image to video” want one thing Runway does not give them, it is that: control over where the shot ends. Our guide to Kling 3.0 free options covers the earlier Kling version.

The honest summary: use Runway image to video AI as the inexpensive baseline, reach for Grok Imagine when you need specific durations or resolutions, PixVerse when you want sound, and Kling when the final frame has to match a specific image. The full model catalog lists the rest.

Is there a free, unlimited image to video AI?

Searches like “image to video ai free unlimited” and “free ai image to video generators” are common, and the honest answer is no, not in a form that holds up for regular use. Video generation runs on expensive GPUs for every second of output, so no provider can give it away without limits.

What you will actually find:

• Free trials. Runway, like most video tools, lets you try the product on a small allowance. It is enough to test the feature, not to run a content pipeline.
• Bundled features. Design suites such as Canva include image-to-video inside their editors. That is convenient if you already pay for the editor, but you get less choice of model and settings than you would calling the models directly.
• “Unlimited” offers. These usually come with low resolution, watermarks, long queues or restrictions on commercial use. Read the terms before building anything on them.

If cost is the real concern, paying per clip is often cheaper than a subscription you underuse. A one-second Grok Imagine Video 1.5 test at 480p is 2.4 credits, rounded up to 3, and Runway is 12 credits per video. Top-up credits never expire, so occasional users are not paying for unused months. See plans and credit prices for the details.

Common mistakes with Runway image to video AI

Most failed clips come from a handful of predictable errors. Fix these and your hit rate improves quickly.

1. Describing the image instead of the motion. The model can see your picture. Spend your words on camera movement, subject action and pace.
2. Asking for motions that reveal the unseen. Orbits and pull-outs force the model to invent the back of objects and the edges of the scene. If you need them, use an image that shows more context, or use a keyframe model with a defined last frame.
3. Using low-resolution or busy start frames. Blur and clutter carry into every frame. Upscale small images first; our AI image upscaler guide covers options.
4. Mismatched aspect ratios. A square image forced into a 16:9 clip gets cropped or padded. Crop the source to the target ratio yourself.
5. Expecting legible text to survive. Labels, signs and logos often warp once they move. Use a static camera for packaging shots, or add text overlays in an editor afterwards.
6. Going long too early. Longer clips drift further from the start frame and cost more on per-second models. Prove the motion at a short duration, then extend.
7. Changing several things at once. Iterate on one variable per run so you know what fixed the problem.
8. Skipping the price check in batches. One wrong resolution setting across hundreds of jobs adds up. Call POST /v1/models/price before you submit a batch.

Runway image to video AI rewards a simple workflow: a clean start frame, a single clear motion, a short first test and small adjustments. Once that works in the playground, the same request moves straight into code, and you can test the identical image on Kling, Grok Imagine or PixVerse by changing one field.