One API for every AI video model

The problem with calling each model directly

Every model maker runs its own API, with its own account, keys, request format, job status flow and billing. Building against three of them means three integrations to maintain, three balances to top up and three sets of error codes to handle. Some models are offered by several platforms at different prices, with different limits, and capacity that comes and goes.

The cost shows up when you want to try a new model, which in 2026 happens every few months. Each new model means another integration before you can even see whether it suits your product.

How Revana works

1. One key. Sign up and create an API key in the dashboard. It works for every model in the catalogue.
2. One request. Send POST /v1/generations with a model id and an input object. The field names are the same for every model: prompt, image_url, end_image_url, image_urls, resolution, duration, aspect_ratio and audio.
3. A price before anything runs. The job's price comes from the catalogue and is held from your prepaid balance. A field or value the model does not support is rejected with a 422 before anything is charged. You can also set max_cost_usd as a ceiling.
4. Routing. The router sends the job to the cheapest route for that model. If the route fails or times out, the job moves to the next one.
5. One result format. Poll GET /v1/generations/{id}. When the status is succeeded, the files are in output.urls and the charge is in cost_usd. If the job fails, the hold is released.

Jobs are asynchronous: images usually take seconds, and video takes from under a minute to several minutes. The docs have the full field reference and error codes.

The same request on three models

Because the request format is shared, comparing models is a loop. This script sends one prompt to Seedance 2.0, Veo 3.1 Fast and Wan 3.0, then prints each result and its cost. Each model keeps its own options, so pick values all three accept; here that is 720p, 8 seconds and 16:9.

import os
import time

import requests

API = "https://api.revana.si/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['REVANA_API_KEY']}"}

PROMPT = (
    "A paper boat sails down a rain-soaked street gutter at dusk, "
    "neon signs reflected in the water. Rain, distant traffic."
)
MODELS = ["seedance-2.0", "veo-3.1-fast", "wan-3.0"]

# Submit the same job to every model; each call returns 202 with an id.
jobs = {}
for model in MODELS:
    resp = requests.post(f"{API}/generations", headers=HEADERS, json={
        "model": model,
        "input": {
            "prompt": PROMPT,
            "resolution": "720p",
            "duration": 8,
            "aspect_ratio": "16:9",
            "audio": True,
        },
        "max_cost_usd": 1.5,
    })
    resp.raise_for_status()
    jobs[model] = resp.json()["id"]

# Poll each job until it finishes.
for model, job_id in jobs.items():
    while True:
        job = requests.get(f"{API}/generations/{job_id}", headers=HEADERS).json()
        if job["status"] in ("succeeded", "failed"):
            break
        time.sleep(5)
    if job["status"] == "succeeded":
        print(model, job["output"]["urls"][0], job["cost_usd"])
    else:
        print(model, "failed:", job["error"])

At October 2026 catalogue prices, the three jobs cost:

The max_cost_usd of 1.5 covers all three. A job that would cost more than its ceiling does not run.

Fallback across routes

A route is one source of upstream capacity for a model. Many models on Revana have more than one, and the router uses them in order:

• Cheapest first. Each job goes to the lowest-cost route that accepts its options.
• Next on failure. If that route fails or times out, the job moves to the next route automatically. The attempts field shows how many were tried.
• Same model, same price. Every route serves the model you asked for, at the listed price. The router never swaps in a different model.
• One charge. Only the route that succeeds is charged. If every route fails, the job ends as failed and costs nothing.

Pricing and credits

Revana is pay as you go. There is no subscription, no seats and no monthly minimum: you buy prepaid credit in packs of $10, $50, $100, $500, $1,000 or $5,000 and spend it on any model at its listed price.

• Most video models are priced per second of output, by resolution and, on some models, by whether sound is on.
• Image models are priced per image, by size.
• The price is held when you submit, charged when the job succeeds and released when it fails.

The pricing page lists every model's prices. Our guides show what common jobs cost on Seedance 2.0, Veo 3.1 and Kling 3.0.

Read the catalogue in code

The full catalogue is a public JSON file at https://revana.si/data/models.json. The main fields for each model:

For example, to list every model that can make a video from text:

# Every model that can make a video from a text prompt
curl -s https://revana.si/data/models.json \
  | jq -r '.models[] | select(any(.tasks[]; . == "text_to_video")) | .id'

Use it to build a model picker, check options before you submit, or estimate a job's price in your own interface. The same data drives the models page.

What to know before you switch

• Options differ by model. The field names are shared, but each model accepts its own values. Kling 3.0, for example, lists no aspect ratio option. Check the model page before you change the model string.
• Output links expire. Download finished files and keep them in your own storage.
• The model is your choice. Routing changes where a job runs, never what model runs it. Falling back to a different model is a decision for your code.
• Prices can change. The catalogue always shows current prices, and each job's price is fixed when you submit it.

Ready to try it? Get an API key, browse the models, or start with the quickstart.