← Models
Realtime video generation
LongLive-2.0 API
A multi-shot clip on main_video; chunk_index resets per scene while session_chunk keeps counting. Served on Windflow as a realtime session you can plug into your product and pay for by the second.
Windflow session capture
Overview
What LongLive-2.0 does
LongLive-2.0 is an autoregressive, multi-shot video diffusion model from NVIDIA Research (NVlabs), built on a Wan2.2-TI2V backbone and designed to keep a long video coherent as the prompt changes. On Windflow you direct it with text shot prompts: switch the shot softly while it runs, or cut to a new scene.
A session keeps the model running: connect once, steer it with text prompts while it runs, and receive the live stream back on a named output track.
Every model in the catalog speaks the same typed session protocol, so switching between models changes the controls you send, not how you integrate.
- Architecture
- Autoregressive multi-shot DiT (Wan2.2-TI2V backbone)
- Resolution
- 1280 × 704
- Chunk
- 29 frames
- Scene budget
- 48 chunks per scene; scene_cut opens a fresh budget
- Inputs
- Text shot prompts only — no reference image in this release
- Outputs
- A multi-shot clip on main_video; chunk_index resets per scene while session_chunk keeps counting
- Delivery
- WebSocket (H.264) · WebRTC
- Session protocol
- sdk.realtime.v2
- Price
- $0.0021 per billable second
$0.0021
Per billable second
Meter starts when generation begins
$7.56
Per billable hour
The same rate, held for an hour
Production
Runtime status
Capacity is brought up on demand
Private beta
Access
Request access or a live demo
Use cases
What teams build with LongLive-2.0
- Multi-shot narrative clips directed live, one text prompt per shot.
- Interactive long-form video where a viewer or an agent decides what happens next.
- Storyboard-to-motion experiments that need scene-to-scene continuity.
- Streams directed scene after scene within the session's published duration limits.
Provenance
Where LongLive-2.0 comes from
The model and its code are published by the upstream authors. Windflow runs it behind one session API and bills per billable second.
- NVlabs/LongLive on GitHubLicence · Apache-2.0
Compare
LongLive-2.0 next to similar models
| Model | Inputs | Output | Price per billable second |
|---|---|---|---|
| LongLive-2.0 | Text shot prompts only — no reference image in this release | A multi-shot clip on main_video; chunk_index resets per scene while session_chunk keeps counting | $0.0021 |
| Helios | Text prompt (required before start) and an optional reference image for image-to-video | One continuous, steerable clip on the main_video track | $0.0017 |
| MiniMax FastH3 | A text prompt with visual, dialogue, soundscape, and music direction | Video on main_video and generated audio downmixed to mono on main_audio, aligned per window | $0.0069 |
| LTX-2.5 | A text prompt and an optional reference still | Video on main_video and the audio generated with it on main_audio, aligned per window | $0.0040 |
FAQ
Questions about LongLive-2.0
- Does LongLive-2.0 accept a reference image?
- Not in this release. It is directed with text shot prompts only; use LTX-2.5 or Helios when you need to start from an image.
- What is the difference between a soft shot change and a scene cut?
- A soft change switches the prompt live and keeps the model's rolling history, so the picture evolves. A scene cut starts a new scene and rebuilds that history, which suits independent content.
- Will a soft shot change preserve every object exactly?
- No. Retaining rolling history is a conditioning policy, not a guarantee of identity or geometry. Evaluate a repeatable shot sequence and inspect continuity as well as the delay until the new direction becomes visible.
- How should I end a directed stream?
- Release the owner session when the experience ends, including error and page-exit paths. Pausing the player does not release model capacity or end billable time after generation has started. Read the session response for its duration limit.
- How much does LongLive-2.0 cost on Windflow?
- LongLive-2.0 costs $0.0021 per billable session-second ($7.56/hr of billable time). Billing starts at the session's first generation_started event and stops when the session is released or ends. Queueing, worker startup, time before generation starts and the documented reconnect grace are not billed. After generation starts, paused or unwatched time remains billable; release the owner session to stop the meter. There is no subscription.
- How do I get access to LongLive-2.0?
- Windflow is in private beta. Request access from the sign-up page, or ask for a live demo of LongLive-2.0. Approved accounts start with $5 and four concurrent sessions.
- Is LongLive-2.0 always running?
- No. Model capacity is brought up on demand rather than kept standing. Live readiness for each model is published at api.windflow.online/v1/status, and a live demo can be arranged on request.
Guides and comparisons
- Live audience use caseBuild a shared AI video live stream for an audience
- GuideChoosing a realtime video model: a decision guide for 12 hosted models
- Model comparisonHelios vs LongLive vs LTX-2.5: choose a realtime video model
- GuideHow to add live AI video to your app and pay per second
- Model API guideHow to use the LongLive 2 API
- Cost comparisonRealtime video API pricing: compare rates and billable timelines
- API guideRealtime video API: generate, edit and control live AI video
- GuideStreaming a video diffusion model to the browser: WebRTC or WebSocket?
- Platform comparisonWindflow vs Reactor: choosing a realtime model session API
- Prompt guideLongLive 2 prompting guide: shots, cuts and timelines
- Use caseInfinite streaming video generation: continuous video that keeps going
- Integration comparisonRealtime video APIs vs batch video generation: which fits?
- GuideHow realtime video models work: autoregressive diffusion, chunks and the latency budget
- Storytelling use caseInteractive AI storytelling with realtime video models
Other models
Realtime video generationHeliosLive video transformationSANA-StreamingLive video transformationEdenLiveLive video transformationJoyAI-Video-EditNavigable world modelLingBot-World 2.0Neural game renderingMatrix-Game 4.0Navigable world modelDreamX-World-5BNavigable world modelABot-World-0-5BRealtime video generationMiniMax FastH3Realtime video generationLTX-2.5Navigable world modelHy-Nomas