WaveSpeed AI API Guide: Pricing, Models & Alternatives (2026)

WaveSpeed AI is a fast hosting platform that runs AI models on demand. It hosts a large catalog of third-party image, video, and audio models, including Nano Banana, Wan, Seedance, and hundreds more, all behind a single API. If you've used Replicate before, the setup will feel familiar: one account, one key, and a long list of models you can call without building a separate integration for each provider. Our Replicate pricing breakdown covers that comparison point in more detail if you're weighing the two directly.

This guide covers what's actually available through the API, real pricing, a working code example, and when a unified API like WaveSpeed AI, or an alternative to it, makes more sense than integrating providers directly.

What Is WaveSpeed AI

WaveSpeed AI describes itself as a fast hosting layer for more than a thousand AI models covering image, video, audio, and 3D generation. Instead of building its own AI models, it hosts models from providers like Google, Alibaba, ByteDance, and Recraft, and offers them all through one consistent request format. You pay WaveSpeed AI directly, and it handles sending your request to whichever underlying model you asked for.

The appeal is simple. Rather than juggling several API keys, several sign-in methods, and several response formats to use several models, you set things up once.

What Models Are Available Through the WaveSpeed AI API

Image models

The catalog includes Google's Nano Banana family (including Nano Banana 2 and Nano Banana Pro), Recraft, Flux 2, Seedream, and a long list of open-weight (models with publicly available weights) and closed-source image models.

Video models

Wan 2.7 and the newer Wan 3.0 reference-to-video model, Seedance 2.0 and its faster variants, and other video generation models sit alongside the image catalog under the same API.

Utility endpoints (upscaling, face swap, background removal)

Beyond generation, WaveSpeed AI also hosts models for common follow-up tasks, including upscaling, background removal, and face swap, so a full pipeline doesn't require leaving the platform.

Every model gets its own model ID in the form vendor/model or vendor/model/variant, for example wavespeed-ai/z-image/turbo, and that ID is what you pass in the request URL.

WaveSpeed AI API Pricing in 2026

WaveSpeed AI charges based on usage. There's no subscription. You add credits to your account and pay per generation. Each model has its own price, and your final cost for a given request depends on factors like output resolution, video duration, and batch size. The exact cost is shown before you submit a generation, either on the model page or through WaveSpeed AI's price-check endpoint.

For open-source models, WaveSpeed AI says its pricing matches what the original provider charges, so you're not paying extra for using the hosting layer. For closed-source models, pricing is set at or below the wider market average. New accounts get $1 in free trial credit to test the platform, though some premium models aren't included. Credits don't expire once purchased.

Larger top-ups also raise your account's rate limits (how many requests you can send per minute, and how many can run at once). WaveSpeed AI runs four account tiers, Bronze, Silver, Gold, and Ultra, with each step up unlocking a higher request limit. A one-time top-up under $1,000 moves you to Silver, and $1,000 to $4,999 gets you Gold. WaveSpeed AI's published tier thresholds for Ultra have varied across its own documentation over time, so confirm the current top-up amount required directly on their pricing page before you plan around it.

Because pricing is set per model rather than as one flat rate, the fair comparison isn't "WaveSpeed AI versus X" in general. It's whichever specific model you plan to call, priced against calling that same model somewhere else. Our pricing calculator guide for unified AI media generation walks through how to set that comparison up.

Authentication and Getting Started

Create a WaveSpeed AI account, generate an API key from your dashboard, and include it as a bearer token (a type of sign-in credential) in every request:

Authorization: Bearer $WAVESPEED_API_KEY

All requests also need a Content-Type: application/json header. From there, submitting a task is a POST request to https://api.wavespeed.ai/api/v3/{model_id}, with a request body that varies depending on which model you're calling.

Working Code Example

Generation on WaveSpeed AI runs in the background by default. You submit a job, get back a prediction ID and a result URL, and check in until it's done. Here's a simple example that generates an image and checks the result.

import os
import requests
import time

api_key = os.environ["WAVESPEED_API_KEY"]
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}

# 1. Submit the task
response = requests.post(
    "https://api.wavespeed.ai/api/v3/wavespeed-ai/z-image/turbo",
    headers=headers,
    json={
        "prompt": "a red circle centered on a white background",
        "size": "1024*1024",
    },
    timeout=(10, 60),
)
response.raise_for_status()
task = response.json()["data"]
task_id = task["id"]
print(f"Task submitted: {task_id}")

# 2. Poll for the result
result_url = f"https://api.wavespeed.ai/api/v3/predictions/{task_id}/result"
poll_interval = 2

while True:
    poll = requests.get(result_url, headers=headers, timeout=(10, 30)).json()
    data = poll["data"]
    if data["status"] == "completed":
        print("Output:", data["outputs"][0])
        break
    if data["status"] in {"failed", "cancelled", "timeout"}:
        raise RuntimeError(data.get("error") or f"Task ended with {data['status']}")
    time.sleep(poll_interval)
    poll_interval = min(10, poll_interval + 1)

Swap the model ID in the submit request for any other model on the platform, and adjust the request body to match that model's own parameters, since the fields differ between an image model and a video model. WaveSpeed AI also supports webhooks (an automatic notification sent to your server when a job finishes) if you'd rather not check in manually, and recommends a polling interval of around 2 seconds for image tasks and 5 seconds for video tasks, increasing the interval for longer-running jobs.

WaveSpeed AI Alternatives: When to Use a Unified API Instead

WaveSpeed AI and similar platforms (Replicate is the other obvious comparison, and fal.ai is another) solve a real problem. Without one, you'd be managing separate sign-ins, error handling, and billing for every model provider you use. But once you've decided to use that kind of platform, the real question is which unified API fits your stack best.

Apiframe covers a lot of the same ground, including several models that overlap directly with WaveSpeed AI's catalog: the full Nano Banana family, Wan image and video models, and Seedance, all behind a single API key and the same request pattern regardless of which underlying model you're calling. If you're already comparing WaveSpeed AI against alternatives, the things worth weighing are model breadth (how many of the specific models you need does each platform host), billing structure (credit-based prepaid balances on both, but check the per-model math for the specific models you'll actually use), and how webhook and polling patterns are implemented, since that affects how much integration work you'll do either way. Our guide to choosing an AI media API covers this decision in more general terms if you're comparing more than two options.

Neither platform is strictly better across the board. The right call usually comes down to which specific models you need and whether you're already using other services from one provider that would simplify billing. Our roundup of AI media generation APIs is a good next read if you want to see how more providers stack up side by side.