Skywork API: Models, Pricing & Integration

There are three different things called “Skywork API”

Searching “skywork api” returns three different products wearing the same name, and only one of them is an API you can call for media generation.

1. Skywork AI, the model lab. The SkyworkAI GitHub org publishes open-source models: the SkyReels video generation family, the SkyClaw agentic language model, vision-language models, and the Skywork Image API. Skywork is the AI lab of Kunlun Wanwei, the Chinese internet company.
2. skywork.ai, the product. skywork.ai is a consumer AI workspace — deep research, documents, slides, and spreadsheets, branded around the “SkyClaw” agent. It is not a model API. There is no “skywork api login” here that gives you an API key for generation models.
3. The API channels. Skywork models are served through access platforms: the official partner channel apifree.ai (SkyReels V3 API, SkyClaw), Runware (SkyReels V4), and aggregators like Modellix that expose the SkyReels family behind one REST API.

A developer asking “is there a Skywork API” is usually asking about number 3: can I send a prompt and get a video back. The answer is yes — through a channel, not through skywork.ai itself. This guide maps the models, the per-second prices as read on August 11, 2026, and the integration path.

Modellix is an AI model API aggregator and has a commercial interest here: we offer 10 Skywork endpoints and are one of the channels described below. The price table comes from our live pages, and the channel comparison states the trade-offs without claiming an aggregator is always the right answer.

Which Skywork models can you actually call by API?

The practical Skywork API surface is the SkyReels video family — ten endpoints on Modellix covering generation, editing, extension, lip-sync, and avatars. SkyReels is Skywork’s video foundation model line: V1 was one of the first open-source human-centric video models, V2 introduced an infinite-length diffusion-forcing architecture, and the SkyReels-V2 repository notes that the SkyReels-V3 API launched on apifree.ai in January 2026.

Endpoint	What it does	Key inputs	Output ceiling
skyreels-t2v	Text to video, optional audio	prompt, duration, resolution	15s, 1080p
skyreels-i2v	Animate one or more keyframes	start/end/mid frames	15s, 1080p
skyreels-r2v	Put 1–4 reference subjects in a new scene	ref images + prompt	5s clips
skyreels-omni	Reference-driven generate/edit (images or video)	ref images/video + prompt tags	video reference ≤15s
video-extension-single-shot	Continue one continuous shot	prefix video + prompt	+5–30s
video-extension-shot-switching	Extend with a new shot or camera angle	prefix video + cut type	+2–5s
video-restyling	Re-render footage into a named style	video ≤30s + style name	720p
sky-lipsync	Re-drive lips to match an audio track	video + audio	720p
single-actor-avatar	Talking avatar from a portrait + audio	portrait image + audio ≤200s	1080p (pro)
segmented-camera-motion	Avatar video with per-segment camera moves	frame + audio + camera control	1080p (pro)

All ten endpoints above, as listed on the live Modellix Skywork provider page, captured August 11, 2026.

The extension and avatar rows slot into larger pipelines: our video extension API guide covers the extension pattern generally, and the talking avatar API guide covers presenter workflows if that is your use case.

Outside the video family there are two more API surfaces worth knowing. SkyClaw-v1.0 is an OpenAI-compatible agent chat API (/agent/v1/chat/completions) for tool-calling and long-context agent workloads. Skywork Image API is documented inside the Skywork-Skills repository — posters, logos, illustrations, and image-to-image editing — and is currently consumed through agent skills rather than a public self-serve console. DeepResearch V2 is also API-accessible, but access is by application rather than self-serve signup.

None of this is “one Skywork API key for everything.” If your product needs video generation, the SkyReels endpoints are the models you want; if you need an agent LLM, SkyClaw is the separate surface.

Skywork API pricing: per-second rates, read from live pages

The ranking results for “skywork api” publish almost no pricing, which is why the numbers below matter: they are read from live model pages on August 11, 2026, and they change. SkyReels video is billed per second of output, with price varying by endpoint, mode, and resolution.

Endpoint	Rate (USD / second)	Mode / resolution notes
skyreels-t2v / skyreels-i2v	$0.0920–$0.4025	fast or std; 480p → 1080p
skyreels-r2v	$0.0805	flat rate
skyreels-omni	$0.1725–$0.7188	fast or std; 480p → 1080p
video-extension-single-shot / video-extension-shot-switching	$0.0690	flat rate
video-restyling	$0.0920	flat rate, 720p
sky-lipsync	$0.0115	flat rate, 720p
single-actor-avatar	$0.0460–$0.0690	std → pro (1080p)
segmented-camera-motion	$0.0920–$0.1150	std → pro (1080p)

Per-second rates at the std 720p baseline where a baseline exists, read from Modellix model pages on August 11, 2026. Flat-rate endpoints are shown at their single listed price.

What the table does not say: a per-second rate is not a per-task cost. A 10-second skyreels-t2v clip at std 720p ($0.1610/s) costs about $1.61; the same endpoint at std 1080p ($0.4025/s) costs about $4.03. Always price the exact duration and resolution you will actually request.

On the official channel, the pricing picture is different: apifree.ai ran SkyClaw free and announced that SkyClaw moves to paid usage on June 8, 2026 with a 25% launch discount for the first week — no per-token rate was published on the page we read. Runware lists SkyReels V4 on its creator page and prices individual generations (example generations on the model page run to about $2.8–$3.5 per clip) rather than publishing a per-second table. For budget comparisons across channels, read the live page of the specific endpoint — this article is a dated snapshot, not a quote. For a sense of where Skywork sits against the broader market, our cheapest AI API roundup covers the low end of the category.

How to integrate: a working SkyReels API call

Media generation is asynchronous: you submit a task, poll for completion, and download the result. The pattern below follows the Modellix REST API documentation and works for any of the 10 Skywork endpoints by changing the model path.

First, get a key. Modellix requires a human-created account; generate an API key in the console and keep it server-side.

curl --request POST \
  --url https://api.modellix.ai/api/v1/skywork/skyreels-t2v/async \
  --header "Authorization: Bearer $MODELLIX_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "prompt": "A lighthouse on a rocky coast at dusk, waves breaking, slow cinematic pan",
    "duration": 5,
    "resolution": "720p",
    "mode": "std"
  }'

The response returns a task_id immediately. Poll it until status is success, then take the output URL from result.resources:

import time
import requests

API = "https://api.modellix.ai/api/v1"
KEY = "your-key-here"  # read from env in production
HEADERS = {"Authorization": f"Bearer {KEY}"}

def generate_video(prompt: str, duration: int = 5, resolution: str = "720p", mode: str = "std") -> str:
    r = requests.post(
        f"{API}/skywork/skyreels-t2v/async",
        json={"prompt": prompt, "duration": duration, "resolution": resolution, "mode": mode},
        headers=HEADERS,
        timeout=30,
    )
    r.raise_for_status()
    task_id = r.json()["data"]["task_id"]

    for _ in range(120):  # 2 minutes of polling
        result = requests.get(f"{API}/tasks/{task_id}", headers=HEADERS, timeout=30).json()["data"]
        if result["status"] == "success":
            return result["result"]["resources"][0]["url"]
        if result["status"] == "failed":
            raise RuntimeError(f"task failed: {result}")
        time.sleep(1)

    raise TimeoutError("video generation timed out")

video_url = generate_video("A lighthouse on a rocky coast at dusk, slow cinematic pan")
print(video_url)  # save promptly: results expire after 7 days

Two practical notes. The skyreels-t2v request takes prompt (max 1280 tokens), duration, resolution (480p/720p/1080p), and mode (fast/std); input-media models like skyreels-i2v take image or video URLs in the request body instead. And results are only stored for 7 days — download to your own storage before then. If you are building an agent workflow rather than a backend service, the Modellix CLI wraps the same task lifecycle in two commands.

Official channel vs aggregators: which Skywork access point fits?

You have three realistic routes to Skywork models, and they are not interchangeable.

Route	What it is	Billing	Trade-offs
apifree.ai	Official partner channel; SkyReels V3 API and SkyClaw	Free tier; SkyClaw moved to paid June 8, 2026	Closest to the source; channel scope and stability are Skywork’s call
Runware	Serverless compute + model API marketplace	Per-generation pricing	Includes SkyReels V4; billing and quotas are Runware’s
Modellix	API aggregator, 12 providers, 210+ media models	Pay-as-you-go per call, no monthly fee, $10–30 free credit	One key and one bill for Skywork plus 11 other providers; same provider-hosted infrastructure

The deciding questions are scope and billing shape. If Skywork is the only model family your product will ever call, the official channel is the natural starting point. If you expect to switch or mix providers — Kling for one workflow, SkyReels for another, Wan for a third — an aggregator removes the per-provider signup, key, and invoice. Modellix’s Skywork provider page lists all 10 currently available endpoints with their live rates; our position in that comparison is not neutral, so treat the aggregator column as the option we sell, and verify the official channel’s current terms before committing.

One more consideration: data residency and support. Skywork is a Chinese lab, and each channel hosts its own infrastructure with its own SLA. If your product has compliance constraints on where generation runs, that constraint selects the channel before price does.

Checklist before you wire Skywork into your product

• Name the Skywork you mean. Lab models, the skywork.ai product, and API channels are three different things; “Skywork API” alone is not a spec.
• Match the endpoint to the task. T2V for scripts, I2V for keyframes, R2V for subject consistency, Omni for edits, avatar endpoints for presenters.
• Price the exact task. Per-second rates vary by mode and resolution; compute duration × rate for the real request shape.
• Confirm the billing unit on your channel. Per-second, per-generation, and free-tier channels do not convert cleanly.
• Decide where generation runs. Channel infrastructure and SLA differ; check compliance constraints before price.
• Download results promptly. Outputs expire (7 days on Modellix) — treat generation as a pipeline, not a mailbox.

FAQ

Is there a free Skywork API?
apifree.ai, the official partner channel, offered free access to Skywork models including SkyClaw; its SkyClaw page announced a move to paid usage on June 8, 2026 with a 25% first-week launch discount. “Free” depends on the channel and the date — check the live page.

Does Skywork have an image API?
Yes. Skywork Image API supports posters, logos, illustrations, and image-to-image editing, and is documented in the Skywork-Skills repository. It is exposed through agent skills today rather than a public self-serve console.

Is skywork.ai an API?
No. skywork.ai is the SkyClaw AI workspace product (deep research, documents, slides, sheets). It does not hand out API keys for media generation models.

How do I get a Skywork API key?
There is no single Skywork API key. For video models, sign up on a channel: apifree.ai (official), Runware, or an aggregator like Modellix, which issues one key covering Skywork and its other providers.

Which models does the Skywork API include?
The accessible surface is the SkyReels video family (text-to-video, image-to-video, reference-to-video, Omni editing, extensions, restyle, lip-sync, avatars), plus the SkyClaw agent chat API and the Skywork Image API. DeepResearch V2 is available by application.

How much does the Skywork API cost?
For SkyReels video on Modellix, $0.0115–$0.7188 per second depending on endpoint, mode, and resolution (read August 11, 2026). SkyClaw was free and moved to paid on June 8, 2026. Rates change; verify on the live page before budgeting.