All models / Video / LongLive 2.0 (real-time multi-shot video)

LongLive 2.0 (real-time multi-shot video)

Name: LongLive 2.0 (real-time multi-shot video)
Brand: NVIDIA
Price: 0.0017 USD
Availability: InStock

model id: reactor/longlive-v2

Real-time · Text in → Video out realtime-video-generation

Hosts

Host	Price	Max session
Reactor	$0.0017 / s	1800 s

0% additional markup beyond our 2.5% platform fee.

No live Playground for this model yet

This model streams from the host’s own SDK, and the browser half of that is yours to build. Switch to the API tab above for the session call, or see Live video & world models for the host’s browser example.

About

LongLive 2.0: autoregressive real-time video built for multi-shot scenes — direct a sequence with soft shot transitions and hard cuts, live or scheduled.

Price

$0.0017 / second

per second the session is live — about $6.27 per hour

Includes the VideoRouter platform fee of 2.5% of provider cost.

Specs

KindReal-time video generation

CreatorNVIDIA

InputText

OutputLive video stream

Resolution704p

Max session1800 s

Default session300 s

How it works

Your server calls POST /v1/live-video/sessions with this model id. We check your key and credits, hold the cost of the session, and return a short-lived credential for the host.
Your browser or app streams from the host directly with that credential — the video never passes through our servers and the host API key never reaches the browser.
Your server sends a heartbeat while it runs and ends the session when the user leaves. You pay per second the session was live, up to 1800 s per session.

Full reference, limits and errors: Live video & world models.

from videorouter_live import LiveVideo
lv = LiveVideo(api_key="llmr_sk_live_...")   # or set VIDEOROUTER_API_KEY

# On YOUR server: open a session, hand `connect` to the browser, settle on exit.
with lv.session(
    "reactor/longlive-v2",
    max_duration_seconds=60,
) as s:
    send_to_browser(s.connect)      # returned once; short-lived credential for the host
    wait_for_user_to_finish()
print(s.status, s.billed_seconds, s.cost_usd)

import { LiveVideo } from "videorouter-live";
const lv = new LiveVideo({ apiKey: process.env.VIDEOROUTER_API_KEY });

// On YOUR server: open a session, hand `connect` to the browser, settle in a finally.
const session = await lv.withSession(
  {
    model: "reactor/longlive-v2",
    maxDurationSeconds: 60,
  },
  async (session) => {
    await sendToBrowser(session.connect);   // returned once
    await waitForUserToFinish();
    return session;
  },
);
console.log(session.status, session.billedSeconds, session.costUsd);

curl https://videorouter.sh/api/v1/live-video/sessions \
  -H "Authorization: Bearer llmr_sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"model": "reactor/longlive-v2", "max_duration_seconds": 60}'

# keep it alive while the user streams, then settle:
curl -X POST https://videorouter.sh/api/v1/live-video/sessions/$SESSION_ID/heartbeat -H "Authorization: Bearer llmr_sk_live_..."
curl -X POST https://videorouter.sh/api/v1/live-video/sessions/$SESSION_ID/end -H "Authorization: Bearer llmr_sk_live_..."