How Kling AI Image to Video Works: Tips and API Code (2026)

Kling AI image to video takes a single still image and turns it into a short moving clip: a portrait that blinks and turns its head, a product that rotates under studio lights, a landscape where clouds drift and water moves. This article covers how the feature works, how to get clean results, what “free” really gets you, and how to run the same image-to-video workflow from code when you need more than a browser tab.

What is Kling AI image to video?

Kling is a generative video service built by Kuaishou, a technology company based in Beijing. It makes videos from text prompts (text-to-video) and from images (image-to-video). Kling AI image to video is the second mode: you give it a picture and it treats that picture as the first frame, then predicts what happens next.

Under the hood, Kling and its competitors are diffusion models. They start from random noise and remove it step by step until a coherent result appears, guided by your inputs. In image-to-video mode, the input image fixes the look of the scene: the subject, the colours, the lighting, the composition. Your text prompt mostly controls motion: what moves, how fast, and what the camera does.

That split explains why image-to-video is so popular. With text-to-video you hope the model draws your character the way you pictured them. With image-to-video you’ve already decided the look, and the model only has to animate it. Designers use it to bring a product render to life, marketers turn a campaign image into a short social clip, and illustrators use it to animate artwork they’ve already finished.

Kling has gone through several model versions, and newer ones generally offer smoother motion, longer clips and higher resolutions. The basic workflow hasn’t changed, and that workflow is the same across almost every image-to-video tool on the market.

How to use Kling AI image to video step by step

The web app and the mobile app follow the same flow. Here is what each step does and what to watch for.

1. Sign in and open the video generator. Pick the image-to-video tab rather than text-to-video. If you can choose a model version, start with the newest one your plan allows.
2. Upload your start image. This becomes frame one. Use the sharpest version you have, and crop it to the aspect ratio you want in the final video (16:9 for YouTube, 9:16 for Reels and TikTok, 1:1 for feeds). Letting the tool reframe the image for you often produces odd padding or stretched edges.
3. Optionally add an end frame. Some Kling versions let you give a last frame too, so the model works out the motion between two fixed images. This is the best way to control where a shot ends up.
4. Write the motion prompt. Describe the movement, not the picture. The model can already see the picture. More on this in the next section.
5. Add a negative prompt if offered. A negative prompt lists things you don’t want, such as “morphing, extra limbs, text, flicker”.
6. Choose length and quality. Longer clips and higher quality modes cost more credits and take longer. Start short.
7. Generate, review, iterate. Expect to run a few versions. Change one thing at a time (prompt wording, length or image crop) so you can tell what helped.

A Kling AI image to video job runs in a queue, so results are not instant. Free users usually wait longer than paid users, and busy periods add more delay.

Writing motion prompts that keep your image intact

Most disappointing results come from prompts that redescribe the image instead of directing it. Think of the prompt as notes to a camera operator and an actor, not a caption.

Separate the subject’s action from the camera move

Write two short clauses: one for what the subject does, one for what the camera does. For example: “The woman slowly turns her head toward the window and smiles. The camera pushes in gently.” Mixing them (“a cinematic smiling woman with dynamic camera”) leaves the model guessing.

Use film vocabulary the models were trained on

• Push in / pull out: the camera moves toward or away from the subject.
• Pan: the camera turns left or right from a fixed position.
• Tilt: the camera turns up or down.
• Orbit: the camera circles the subject.
• Static shot: the camera stays still and only the subject moves. Useful when you need the frame to stay clean, as with logos and products.

Ask for less motion than you think

A photo holds information about one moment from one angle. Every big movement forces the model to invent things it can’t see: the back of a head, the other side of a product, a hand that was out of frame. Small, slow movements stay faithful to the source, while large, fast ones drift. “Hair moves slightly in the breeze, she blinks” will look far more real than “she spins around and dances”.

A worked prompt

Source image: a coffee cup on a wooden table by a window, morning light.

• Weak: “A beautiful cup of coffee on a table, cinematic, 4k, amazing.”
• Strong: “Steam rises slowly from the coffee. Sunlight flickers softly through the window. Slow push in toward the cup. Static table, no objects move.”

The strong version names three motions, gives their speed, and says what must stay still. That last part matters: telling the model what should not move is one of the most reliable ways to stop objects melting into each other.

Is Kling AI image to video free, and what about “unlimited”?

People searching for Kling AI image to video free, or for “image to video AI free unlimited”, usually want to know if they can skip paying entirely. The honest answer: you can try it for free, but you can’t use it at volume for free.

Kling has a free tier with a limited credit allowance. It’s enough to test the feature and make a few clips. Paid plans add more credits, faster queues and access to higher quality settings. The exact allowances change often, so check the plan page on Kling’s own site before you rely on them.

Lists of free AI image to video generators tend to show the same pattern: a small daily or monthly allowance, watermarks or lower resolution on free output, and long queues. There’s a simple reason. Every second of generated video takes serious GPU time, so a service offering truly unlimited free video either has a catch (heavy watermarks, very low resolution, data use you might not accept) or won’t last.

Design suites have also added the feature. Canva AI image to video and similar built-in tools are handy if you already work inside that editor and only need the odd clip. They give you less control over the underlying model, and you can’t call them from your own code.

If you’re past the testing stage, the sensible question is what each clip costs and how predictable that cost is. That’s where an API with clear per-clip pricing helps.

Running image to video through an API

A web app works well when you make a handful of clips by hand. It doesn’t help when you want to animate 200 product photos, add image-to-video to your own app, or let a coding agent produce clips for you. For those jobs you need an API.

Unicorn API gives you one key and one bill for many video models, priced in credits. Kling AI image to video is one option among several models built for the same task, and the ones below all take a start image and a motion prompt. You can compare them side by side in the browser playground before writing any code.

Model	Lengths	Resolutions	Price (credits)	Good for
Grok Imagine Video 1.5	1-15 s	480p, 720p, 1080p	3 / 4 / 7 per second	Cheapest per second, strong motion; image-to-video only
Wan 3.0	2-30 s	480p, 720p, 1080p	6 / 12 / 24 per second	Native synced audio included, optional end image
Vidu Q4 Image To Video	Set by duration	540p up to 4k	7.2 (540p) to 62.4 (4k) per second	When you need 2k or 4k output
Wan 2.6 Flash Image To Video	5, 10, 15 s	720p, 1080p	20 (5 s, 720p, no audio) to 180 (15 s, 1080p, audio)	Fixed-length clips with an audio toggle
Grok Imagine Video 1.0	6-30 s	480p, 720p, 1080p	3 / 4 / 7 per second	Longer clips at the low Grok rate

On a monthly plan, a credit costs $0.01. A 5-second 720p clip on Grok Imagine Video 1.5 is 20 credits, so $0.20. The same clip at 1080p is 35 credits. A 10-second 720p clip on Wan 3.0, with audio included, is 120 credits. Top-up credits cost $0.015 each and never expire. Plans run from Starter at $25 to Pro at $200 a month; see pricing for the details.

You never have to guess: POST /v1/models/price returns the exact price of a request before you run it, which helps when your app lets users pick length and resolution. Request formats are in the API documentation.

How a video job works

Video generation takes too long for a single HTTP request, so it runs as an asynchronous job. You POST /v1/jobs with the model id and inputs, get back a job id, then poll GET /v1/jobs/{id} until the status is "succeeded". The video URL is in outputs[0].url.

Worked example: animating a product photo with Python

Say you run a shop and want a 5-second vertical clip of each product for social media. Here’s a complete script using Grok Imagine Video 1.5, the cheapest per second of the models above. Create a key on the API keys page and export it as UNICORN_API_KEY first.

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {
    "Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}",
    "Content-Type": "application/json",
}

def image_to_video(image_url, prompt, duration=5, resolution="720p", aspect_ratio="9:16"):
    body = {
        "model": "grok-imagine-video-1-5-preview",
        "input": {
            "image": image_url,
            "prompt": prompt,
            "duration": duration,
            "resolution": resolution,
            "aspect_ratio": aspect_ratio,
        },
    }
    job = requests.post(f"{API}/v1/jobs", json=body, headers=HEADERS).json()
    job_id = job["id"]

    while True:
        job = requests.get(f"{API}/v1/jobs/{job_id}", headers=HEADERS).json()
        if job["status"] == "succeeded":
            return job["outputs"][0]["url"]
        if job["status"] == "failed":
            raise RuntimeError(f"Job {job_id} failed: {job}")
        time.sleep(5)

url = image_to_video(
    "https://example.com/products/ceramic-mug.jpg",
    "The mug rotates slowly on the table. Soft studio light glints on the glaze. "
    "Static camera, plain background, nothing else moves.",
)
print("Video ready:", url)

This job costs 20 credits (5 seconds at 720p). For a catalogue of 200 products, that’s 4,000 credits, or $40 on a monthly plan. You know that figure before you start, which is hard to say for credit systems that vary by queue or quality mode.

Controlling the end frame with Wan 3.0

When you need the clip to finish on a specific shot, such as a packshot with the logo facing forward, Wan 3.0 takes an end_image as well as the start image. It also generates synced audio at no extra cost. Here is the same idea in curl:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan/3-0-video",
    "input": {
      "image": "https://example.com/products/mug-side.jpg",
      "end_image": "https://example.com/products/mug-front.jpg",
      "prompt": "The camera orbits the mug from the side to the front. Soft ambient cafe sound.",
      "duration": 5,
      "resolution": "720p",
      "aspect_ratio": "9:16"
    }
  }'

That request is 60 credits (5 seconds at 720p, audio included). Poll it exactly as in the Python example. If you prefer to work from an editor, the same models are available to Cursor, Claude Code and other coding agents through MCP.

For longer, character-driven animation, our guide to the AI animation video generator options compares more models. If you specifically want Kling’s own motion tools, we cover Kling 2.6 Motion Control separately.

Common image to video mistakes and how to fix them

These problems come up whether you use Kling AI image to video in the browser or call a model through an API.

• Using a low-quality source image. Blur, heavy JPEG compression and noise get amplified, not fixed. Start from the sharpest file you have. If the image is AI-generated, fix warped hands or garbled text first, because the video will keep them.
• Wrong aspect ratio. A landscape photo forced into a 9:16 video gets padded or cropped badly. Crop the image to the target ratio before uploading.
• Asking for too much motion. Big actions cause faces to drift and objects to melt. Ask for slower, smaller movements and fewer of them.
• Redescribing the image. “A red car on a road” adds nothing the model can’t see. Say what moves: “The car drives slowly toward the camera, dust rises behind it.”
• Going long too early. Long clips cost more and drift more. Get a 5-second version right, then extend. To build a longer sequence, use the last frame of one clip as the start image of the next.
• Not saying what stays still. “Background static, logo does not move” heads off a lot of strange artefacts.
• Paying top resolution on drafts. Test prompts at 480p (3 credits per second on Grok Imagine Video 1.5), then render the final at 1080p. A draft costs less than half as much, so you can iterate more.
• Not handling failures in code. Jobs can fail or queue for a while at peak times. Grok Imagine Video 1.5 in particular can be busy. Always check for a failed status, add a timeout, and keep a second model as a fallback.

Kling AI image to video is a good place to learn the craft: one image, a clear motion prompt and a short clip. Once you need volume, automation or predictable costs, the same skills carry straight over to an API. Browse the full model catalog to compare every video model on one bill.