GPT Image 2.5 is live — OpenAI's newest image model, targeted edits that leave the rest of the frame alone
All Posts
Image to Video API in 2026: Prices, Inputs, and Free Limits
2026/08/14

Image to Video API in 2026: Prices, Inputs, and Free Limits

Image to video API routes compared: source-image, reference, and first/last-frame inputs across six models, with live per-second prices from $0.03/s.

If you are looking for an image to video API to automate video creation from stills, the price range in August 2026 runs from $0.03 per second of output to $0.18 and up, and the bigger differences are in what "image to video" means to each model. Some APIs animate one source image. Some treat your images as identity references and generate a new scene around them. One accepts a first and a last frame and fills the motion in between.

Those are three different jobs, and picking a model by price before picking the input shape is how automation pipelines end up rebuilt twice. This guide sorts six models by input shape first, then puts live per-second rates and an automation script against each, with every number pulled from the models' pages on reAPI in August 2026.

TL;DR

  • Cheapest way to animate a still, and the cheapest first-and-last-frame route: Seedance 2.0 Mini at $0.049875/s for 480p image-only generation[2]. Its $0.0315 source-video rate does not apply to images or first/last frames.
  • Simplest single-image route: Grok Imagine Video 1.5 takes exactly one source image and runs $0.03/s at 480p on its official tier, $0.06/s at 720p[1].
  • Most reference room: Seedance 2.5 accepts up to 30 reference images per request[4]; MiniMax H3 takes 9 plus a first and last frame[5].
  • First-and-last-frame control is the rare feature: of the six, only MiniMax H3 and Seedance 2.0 Mini expose both endpoints of the clip as inputs[2][5].
  • No API on this list offers unlimited free generation; video inference is too expensive to give away. Signup credits exist for testing, and per-second billing means a real budget stays under a dollar per test clip.

Three input shapes hiding behind "image to video"

Source-image animation. The model receives one image and generates motion from it: the picture is frame one, or close to it. Grok Imagine Video 1.5 is the pure version of this shape; its image_urls field is required and takes exactly one image[1]. This is the shape people usually mean by "make my photo move."

Reference-driven generation. The images steer identity, product, or style while the model composes a new scene. Seedance 2.5 takes up to 30 reference images, Kling 3.0 and Vidu Q3 take up to two[4][3][6]. Character consistency across a series of clips lives here. On Seedance 2.5, image references stay on the $0.118589/s 480p base rate; only a source video selects the $0.071153 rate, and its duration also becomes billable[4].

Frame interpolation. Supply the first frame, optionally the last, and the model connects them. MiniMax H3 exposes first_frame_url and last_frame_url together with up to nine reference images[5], and Seedance 2.0 Mini carries the same pair of frame fields plus nine image references at a fraction of the rate[2]. This is the control-heavy shape for storyboarded shots.

Six image to video APIs at their live rates

All six run behind the same endpoint on reAPI; rates below are the listed per-second prices retrieved in August 2026.

ModelInput shapeImage slotsClip lengthLowest listed rate
Seedance 2.0 Minifirst/last frame + refs9 + 2 frames4–15 s$0.049875/s at 480p[2]
Vidu Q3referenceup to 21–16 s$0.0368/s at 540p turbo[3]
Grok Imagine Video 1.5source image (required)11–15 s$0.03/s at 480p[1]
MiniMax H3first/last frame + refs9 + 2 frames4–15 s$0.1125/s at 768P, audio included[5]
Kling 3.0referenceup to 23–15 s$0.077/s, no audio[6]
Seedance 2.5referenceup to 304–30 s$0.118589/s at 480p with image input[4]

Resolution ceilings differ: Grok tops out at 720p on this surface[1], Vidu Q3 lists 1080p tiers up to $0.1472/s[3], Kling 3.0 has a 4K tier at $0.3685/s[6]. Every model here rejects base64 uploads; images go in as public HTTPS URLs.

What eight seconds costs from one image

The same eight-second job, cheapest applicable tier on each model:

Model, tier8-second clip
Grok Imagine Video 1.5, official 480p$0.24[1]
Vidu Q3, turbo 540p$0.29[3]
Seedance 2.0 Mini, 480p with image input$0.304[2]
Seedance 2.5, 480p with image reference$0.949[4]
Kling 3.0, standard no audio$0.62[6]
MiniMax H3, standard with audio$1.46[5]

The spread is about 6× between the cheapest tier and the most expensive, which is why the input shape decision comes first. Paying H3 rates to animate a single product photo buys control you are not using; feeding one still into a reference-driven model and expecting frame-one fidelity produces a different scene than you meant.

Automating the pipeline in Python

Automation is a loop: submit one task per image, poll each task id, collect output URLs. Credits are prepaid at 1 credit = $0.001 and failed tasks refund automatically, so a bad URL in the batch costs nothing[1].

import requests, time

API = "https://reapi.ai/api/v1"
HEADERS = {"Authorization": "Bearer rk_live_..."}

images = [
    "https://your-cdn.com/product-01.png",
    "https://your-cdn.com/product-02.png",
]

tasks = []
for url in images:
    r = requests.post(f"{API}/videos/generations", headers=HEADERS, json={
        "model": "grok-imagine-video-1.5-beta",
        "image_urls": [url],
        "prompt": "Slow 180-degree orbit around the product, studio lighting",
        "resolution": "480p",
        "duration": 8,
    }, timeout=30)
    tasks.append(r.json()["id"])

for task_id in tasks:
    while True:
        poll = requests.get(f"{API}/tasks/{task_id}", headers=HEADERS).json()
        if poll["status"] in ("completed", "failed"):
            print(task_id, poll["status"], poll.get("output") or poll.get("error"))
            break
        time.sleep(2)

Two things to respect in production: keep total request rate under the platform's 5 requests/second limit including polling, and stage source images on a host that serves fast, since the model fetches each URL server-side.

The "free unlimited" question

A recurring search is an image to video AI that is free and unlimited. It does not exist as stated, on any platform, because video inference occupies expensive accelerators for the full render time; anything marketed as unlimited meters you somewhere else, through queues, throttles, watermarks, or resolution caps. What does exist: signup credits with no card required, enough to run real test calls, and per-second billing after that. At the rates above, an eight-second 480p test costs about a quarter. Testing every model in this table on your own footage costs about $3.86, which is a defensible line item for picking the engine behind an automated pipeline.

FAQ

Can you recommend APIs for automating video creation from images?

Match the API to the job: Grok Imagine Video 1.5 for animating single stills at $0.03/s[1], Seedance 2.5 for reference-heavy character or product work[4], MiniMax H3 when you need to pin the first and last frame[5]. All of them run through one endpoint on reAPI, so the automation code stays identical while you switch models.

Is there a free unlimited image to video AI?

No. GPU time makes unlimited free video generation economically impossible, and offers using that phrasing meter through queues or output limits instead. Free signup credits for testing plus per-second rates from $0.049875/s is the honest shape of the market[2].

Which image to video API is cheapest?

For a job driven only by images, Seedance 2.0 Mini starts at $0.049875/s, Vidu Q3 Turbo at $0.0368/s, and Grok Imagine Video 1.5 at $0.03/s[1][2][3]. Mini's lower $0.0315 rate requires an actual source video and bills source plus output duration, so it is not the image-to-video price.

Can I control both the first and last frame?

MiniMax H3 and Seedance 2.0 Mini both accept first_frame_url and last_frame_url in one request[2][5]. Those are the tools for storyboarded sequences where clip N's end must match clip N+1's start.

How many reference images can I send?

Seedance 2.5 takes up to 30, MiniMax H3 and Seedance 2.0 Mini up to 9 each, Kling 3.0 and Vidu Q3 up to 2, Grok Imagine Video 1.5 exactly 1[1][3][4][5][6]. More slots cost more per second, so send the minimum set that locks the identity you need.

Do image to video outputs include audio?

MiniMax H3 always generates audio at its flat rate; Seedance 2.5 ships audio on by default; Kling 3.0 prices audio and no-audio tiers separately[4][5][6]. Grok Imagine Video 1.5 and Vidu Q3 price by resolution tier on this surface.

Picking an input shape in practice

Decide what the image is: the first frame, a reference, or one of two endpoints. That single decision eliminates most of the table. Then let price sort the survivors, and spend one batch of signup credits running your own stills through the two finalists before wiring either into production. An image to video API earns its place in a pipeline by behaving the same on image five hundred as on image one, and at $0.24 per test clip, proving that costs less than one wrong afternoon of rework.

References

  1. reAPI. Grok Imagine Video 1.5 — model page with live pricing and input schema. Retrieved August 2026 from reapi.ai/models/grok-imagine-video-1-5
  2. reAPI. Seedance 2.0 Mini — model page with live pricing and source-video billing. Retrieved August 17, 2026 from reapi.ai/models/seedance-2-0-mini
  3. reAPI. Vidu Q3 — model page with live pricing and input schema. Retrieved August 2026 from reapi.ai/models/viduq3
  4. reAPI. Seedance 2.5 — model page with live pricing and input schema. Retrieved August 2026 from reapi.ai/models/seedance-2-5
  5. reAPI. MiniMax H3 — model page with live pricing and input schema. Retrieved August 2026 from reapi.ai/models/minimax-h3
  6. reAPI. Kling 3.0 — model page with live pricing and input schema. Retrieved August 2026 from reapi.ai/models/kling-3-0

Further reading

All Posts

More Posts