# Windflow > Windflow runs realtime interactive AI video models as on-demand sessions. Teams plug them into their products and pay by the session second. Windflow (Windflow Labs, https://windflow.online) is a developer platform for realtime interactive video and world models. A session keeps the model running: connect once, receive live frames, and send prompts, images, camera motion or actions over the same channel while the model keeps running. The API is REST for control (`/v1`) plus a media transport (WebSocket or WebRTC). Windflow is in private beta. Windflow Labs is not affiliated with other organisations that use the Windflow name: the AI-powered QA and UX testing product at windflow.io, the enterprise AI company at windflow.ai, the WindFlow C++ stream-processing library, or Windflow Technology, the wind-turbine company. ## Models 12 models are in the public catalog. Each speaks the same typed session protocol; what differs is the inputs and controls a model supports. - [Helios](https://windflow.online/models/helios): One continuous, steerable clip on the main_video track. Inputs: Text prompt (required before start) and an optional reference image for image-to-video. Architecture: 14B Diffusion Transformer, autoregressive chunked diffusion. $0.0017 per billable session-second. - [LongLive-2.0](https://windflow.online/models/longlive-2): A multi-shot clip on main_video; chunk_index resets per scene while session_chunk keeps counting. Inputs: Text shot prompts only — no reference image in this release. Architecture: Autoregressive multi-shot DiT (Wan2.2-TI2V backbone). $0.0021 per billable session-second. - [SANA-Streaming](https://windflow.online/models/sana-streaming): A source-guided edited stream on main_video. Inputs: A source clip you upload (≥ 41 frames) or a live webcam track, plus an edit prompt. Architecture: 2B streaming editing Diffusion Transformer. $0.0019 per billable session-second. - [EdenLive](https://windflow.online/models/edenlive): A proxy-guided generated video stream on main_video. Inputs: Source game footage or a live video track, a scene prompt, and an optional visual reference. Architecture: 5B autoregressive diffusion transformer with proxy-DINO control and LightTAE decoding. $0.0021 per billable session-second. - [JoyAI-Video-Edit](https://windflow.online/models/joyai-video-edit): A source-guided edited stream on main_video. Inputs: A source clip you upload or a live webcam track, an edit instruction, and optionally a reference image. Architecture: 16B multimodal DiT, chunk-autoregressive, two-step distilled. $0.0021 per billable session-second. - [LingBot-World 2.0](https://windflow.online/models/lingbot): A navigable stream on main_video you steer in real time. Inputs: A seed image (uploadFile) and a text prompt, plus persistent WASD movement and look controls. Architecture: Seed-image conditioned navigable world model. $0.0024 per billable session-second. - [Matrix-Game 4.0](https://windflow.online/models/matrix-game-4): The live game, re-rendered by the scene model, on main_video. Inputs: A scene choice and held keyboard controls; no prompt or seed image. Architecture: Scene-specific 5B causal video renderers (Wan2.2-Fun-5B-Control base, one denoising step) driven by an executable game. $0.0024 per billable session-second. - [DreamX-World-5B](https://windflow.online/models/dreamx-world-5b): A controllable world rollout on main_video. Inputs: A seed image, a text prompt, and movement/camera-control state. Architecture: 5B controllable world generation model (Wan2.2-TI2V backbone). $0.0023 per billable session-second. - [ABot-World-0-5B](https://windflow.online/models/abot-world-0-5b): A third-person navigable world stream on main_video. Inputs: A seed image, a text prompt, and persistent movement/camera key state. Architecture: 5B keyboard-conditioned world model (Wan2.2-TI2V backbone, LongForcing causal student). $0.0023 per billable session-second. - [MiniMax FastH3](https://windflow.online/models/fast-h3): Video on main_video and generated audio downmixed to mono on main_audio, aligned per window. Inputs: A text prompt with visual, dialogue, soundscape, and music direction. Architecture: 35B MiniMax H3 text-to-video-and-audio model, DMD2-distilled to four DiT calls with 90% sparse VSA-H3 attention. $0.0069 per billable session-second. - [LTX-2.5](https://windflow.online/models/ltx-2-5): Video on main_video and the audio generated with it on main_audio, aligned per window. Inputs: A text prompt and an optional reference still. Architecture: 22B joint audio+video DiT, two-stage, generated one window at a time. $0.0040 per billable session-second. - [Hy-Nomas](https://windflow.online/models/hy-nomas): Silent generated video on main_video. Inputs: Text and persistent camera movement/look snapshots. Architecture: Distilled Hy-Nomas DiT with a stable-prefix PixelUMM decoder. $0.0047 per billable session-second. ## Pricing Pay per billable session-second, from $0.0017 to $0.0069 depending on the model (indicative private-beta rates). 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. No subscription, no seats, no minimums. Approved accounts start with $5 and 4 concurrent sessions. Details: https://windflow.online/pricing. ## Capacity Model capacity is brought up on demand. Live readiness per model is at https://api.windflow.online/v1/status (no auth). To see a model running, ask for a live demo. ## World model directory - [World model directory](https://windflow.online/world-models): 56 source-linked entries; updated 2026-11-03. Directory entries include models Windflow does not host. - [Open-weight world models](https://windflow.online/world-models/open-weights): Interactive world models with downloadable weights, with the license, controls and real-time claim for each, linked to primary sources. - [World models with permissive licenses](https://windflow.online/world-models/commercial-use): World models whose weight model cards declare a permissive license such as Apache-2.0 or MIT, with the base-model caveats that still apply. - [Real-time world models](https://windflow.online/world-models/real-time): World models whose authors say they generate fast enough to interact with live, with each author's own figures and sources. - [World models you can try online](https://windflow.online/world-models/try-online): World models you can use without setting them up: official demos and APIs, plus hosting platforms, listed alphabetically. - [Game world models](https://windflow.online/world-models/games): World models that turn keyboard, mouse or controller input into game frames, with weights, licenses and places to play. - [Explorable world models](https://windflow.online/world-models/explorable-worlds): World models that turn an image or prompt into a place you can move through with keys, camera paths or actions. - [3D world builders](https://windflow.online/world-models/3d-world-builders): World models that build a persistent 3D scene from text, images or video that you can explore, edit or export, instead of streaming video frames. - [Live, steerable video models](https://windflow.online/world-models/live-video): Models that keep generating video while you change the prompt, the source footage or the scene, instead of rendering one fixed clip. - [Driving world models](https://windflow.online/world-models/driving): World models that predict a vehicle's front-camera view from past frames, some steered by a trajectory, with weights, licenses and demos. - [World models for robotics and driving](https://windflow.online/world-models/robotics-and-driving): Action-conditioned world models that predict what a robot or vehicle sees next, for testing and training policies in simulation. - [Open dataset and corrections](https://github.com/windflow-labs/world-model-directory): source data, change history and contribution instructions. ## Radar - [Realtime models radar: week of October 5, 2026](https://windflow.online/radar/2026-10-11-realtime-models-radar): What shipped or moved in realtime video and world models this week, each with its source: Odyssey-3, Vivix W1, Reactor's ComfyUI nodes, TaoMate-H3, FastH3 V2 and LingBot-Map. ## Guides - [Choosing a realtime video model: a decision guide for 12 hosted models](https://windflow.online/guides/choosing-a-realtime-video-model): Pick between 12 hosted realtime video and world models by inputs, live controls, audio, transports, runtime status and current API prices. - [EdenLive compute measurements on H100, MI355X and RTX PRO 6000](https://windflow.online/guides/edenlive-gpu-compute-measurements): Compare six archived EdenLive model-loop measurements on H100, MI355X and RTX PRO 6000, with downloadable data and explicit browser-latency limits. - [How per-second pricing works for realtime video, with worked examples](https://windflow.online/guides/how-per-second-pricing-works): Understand billable-second video API pricing, compare model rates and estimate session costs. Queueing and startup are free; release sessions to stop billing. - [How to add live AI video to your app and pay per second](https://windflow.online/guides/add-live-ai-video-to-your-app): Add live AI video to your product: choose a model, connect a session, stream frames, send controls and estimate billable session costs with Windflow. - [How to measure realtime video API latency and responsiveness](https://windflow.online/guides/measure-realtime-video-latency): Measure first-frame time, control-to-visible delay, delivered FPS and stalls. Evaluate a realtime video or world model through the client your users see. - [How to use the Helios API](https://windflow.online/guides/helios-api): Use the Helios API from a browser: mint a token on your server, wait for ready, stage the opening prompt, draw the live stream and release the session. - [How to use the LingBot API](https://windflow.online/guides/lingbot-api): Use the LingBot API to stream a navigable world: seed image and prompt in one call, WASD as ordered key snapshots, events mid-run and a clean release. - [How to use the LongLive 2 API](https://windflow.online/guides/longlive-2-api): Use the LongLive-2.0 API to stream a multi-shot video: stage the opening shot, start, then steer with soft shots, hard cuts and shots scheduled ahead of time. - [How to use the SANA-Streaming API](https://windflow.online/guides/sana-streaming-api): Edit a live webcam or an uploaded clip with the SANA-Streaming API: one WebSocket for camera and video, an edit prompt, periodic re-anchoring and release. - [JoyAI live editing latency: what the 443 ms measurement includes](https://windflow.online/guides/joyai-live-editing-latency-measurements): Read JoyAI's 443 ms result with its loopback measurement boundary. Compare frame processing and emission stages with downloadable data and sample counts. - [Realtime video API: generate, edit and control live AI video](https://windflow.online/guides/realtime-video-api): Choose a realtime video API for continuous generation, webcam editing or interactive worlds. Compare controls, latency, session lifecycle and usage costs. - [SANA camera delivery: why 25 fps can still mean a slow response](https://windflow.online/guides/sana-camera-fps-and-latency): Six SANA camera runs separate native FPS, interpolated playback, buffering and visible response. Download the November 7 measurements and method. - [Streaming a video diffusion model to the browser: WebRTC or WebSocket?](https://windflow.online/guides/stream-video-diffusion-webrtc-websocket): Deliver streaming video diffusion to the browser with WebRTC or WebSocket. Understand generation chunks, network trade-offs and visible latency. - [World model API: embed a navigable AI world in your app](https://windflow.online/guides/world-model-api): Embed navigable AI worlds with a hosted world model API. Compare LingBot, Matrix-Game, DreamX and ABot inputs, movement controls and integration limits. - [Helios prompting guide: write and steer a live video](https://windflow.online/guides/helios-prompting-guide): Write Helios prompts that hold up in a live stream: the opening prompt, soft steering versus scene cuts, scheduled beats, reference images and image strength. - [How to use the DreamX World API](https://windflow.online/guides/dreamx-world-api): Use the DreamX-World-5B API to roll a seed image forward as a world you steer: stage prompt and seed together, send movement and look state, and release. - [How to use the EdenLive API](https://windflow.online/guides/edenlive-api): Render a game or live video with the EdenLive API: send greybox frames up one WebSocket, set a scene prompt and visual anchor, and draw the generated stream. - [How to use the Hy-Nomas API](https://windflow.online/guides/hy-nomas-api): Stream a Hy-Nomas world from a text prompt: open it with set_prompt, steer it with held keys and text events, and plan each world as a 60-second take. - [How to use the JoyAI Video Edit API](https://windflow.online/guides/joyai-video-edit-api): Edit a live webcam or an uploaded clip with the JoyAI-Video-Edit API: server token, one WebSocket for camera and video, the edit instruction and release. - [How to use the LTX-2.5 API](https://windflow.online/guides/ltx-2-5-api): Stream LTX-2.5 video with generated audio: join the session over WebRTC, play picture and sound in one element, steer window by window, and release. - [How to use the Matrix-Game 4.0 API](https://windflow.online/guides/matrix-game-4-api): Stream a playable game re-rendered by Matrix-Game 4.0: pick a scene before start, send held-key snapshots on the socket, and plan for 60-second sessions. - [How to write edit instructions for realtime video editing](https://windflow.online/guides/realtime-video-edit-instructions): Write edit instructions that change only what you mean in a live webcam or clip: instructions not captions, what changes then what stays, and reference images. - [LongLive 2 prompting guide: shots, cuts and timelines](https://windflow.online/guides/longlive-2-prompting-guide): Write LongLive-2.0 shot prompts for a live multi-shot video: one shot per prompt, soft shots versus hard cuts, continuity anchors and a timeline planned ahead. - [LTX-2.5 prompting guide: scenes, sound and dialogue](https://windflow.online/guides/ltx-2-5-prompting-guide): Write LTX-2.5 prompts for a live stream with sound: one audio-visual brief per ten-second window, dialogue in quotes, continuity anchors and when to cut. - [World model seed images and prompts: a practical guide](https://windflow.online/guides/world-model-seed-images-and-prompts): Pick seed images and write prompts that keep a navigable world model stable: matching image and prompt, framing for movement, world events and control rhythm. - [How to turn an open-source video model into a realtime API](https://windflow.online/guides/turn-an-open-source-video-model-into-a-realtime-api): Prepare a video model for a hosted realtime API: licensed weights, repeatable inference, streaming transport, session controls and measured performance. - [What is a world model? How it differs from video generation](https://windflow.online/guides/what-is-a-world-model): A world model predicts how an environment changes as you act in it. What that means, how it differs from video generation, what people use it for, and where it falls short. - [How to recover a realtime video stream after a dropped connection](https://windflow.online/guides/recover-a-dropped-realtime-video-stream): Tell a dropped connection from an ended session, retry inside a bounded grace, reattach to the same session and restore playback, with a test checklist. - [How to use the ABot World API](https://windflow.online/guides/abot-world-api): Use the ABot-World API to stream a third-person world: WebSocket-only delivery, seed image and prompt staged together, eight-key controls and release. - [Realtime AI video on AMD GPUs: reading MI355X benchmarks](https://windflow.online/guides/realtime-video-generation-amd-gpus): Read Windflow's AMD MI355X video inference measurements with the right context: GPU count, resolution, generated windows, edit latency and runtime limits. - [How realtime video models work: autoregressive diffusion, chunks and the latency budget](https://windflow.online/guides/how-realtime-video-models-work): Why standard video diffusion cannot run live, what autoregressive chunked generation and few-step distillation change, where the latency goes, and how to read a realtime claim. ## Comparisons - [Helios vs LongLive vs LTX-2.5: choose a realtime video model](https://windflow.online/compare/helios-vs-longlive-vs-ltx): Choose Helios for continuous scenes, LongLive for shot direction or LTX-2.5 for video with sound. Compare control timing and inputs before integration. - [Managed platform, rented GPU or self-hosting: how to run a realtime video model](https://windflow.online/compare/managed-platform-vs-rented-gpu-vs-self-hosting): Compare managed APIs, rented GPUs and self-hosting for realtime video models. Understand session costs, infrastructure work and when each approach fits. - [Realtime video API pricing: compare rates and billable timelines](https://windflow.online/compare/realtime-video-api-pricing): Compare dated Windflow, Reactor, Decart and fal realtime rates. Estimate billable session time, distinguish pricing units and account for retained idle time. - [Windflow vs Decart: evaluating a live video editing API](https://windflow.online/compare/windflow-vs-decart): Evaluate Decart Lucy and Windflow SANA or JoyAI for live camera editing. Compare reference inputs, media paths, token limits and measurement boundaries. - [Windflow vs fal: choosing the right realtime video endpoint](https://windflow.online/compare/windflow-vs-fal): Compare Windflow sessions with fal's job, streaming and realtime endpoints. Understand Lucy WebRTC signaling, model controls and migration work. - [Windflow vs Reactor: choosing a realtime model session API](https://windflow.online/compare/windflow-vs-reactor): Compare Windflow and Reactor by session lifecycle, model-specific controls, billing boundaries and integration work, using dated primary sources. - [Realtime video APIs vs batch video generation: which fits?](https://windflow.online/compare/realtime-video-api-vs-batch-generation): Choose live sessions or completed video jobs by interaction, latency and cost. Understand streaming output, state, controls and when to use both patterns. ## Use cases - [Build a shared AI video live stream for an audience](https://windflow.online/use-cases/shared-ai-video-live-stream): Stream one generated AI video session to authorized viewers. Design host controls, audience input, session ownership, billing and recording boundaries. - [Realtime game rendering: restyle a game live with AI](https://windflow.online/use-cases/restyle-a-game-live-with-ai): Let your game run the rules and a realtime model draw the frames: how neural game rendering works, EdenLive versus Matrix-Game 4.0, and how to write the scene. - [Infinite streaming video generation: continuous video that keeps going](https://windflow.online/use-cases/infinite-streaming-video-generation): Build continuous AI video streams with hosted realtime models. Compare prompt control, generation windows and session limits for channels and displays. - [Live video editing from a webcam with a text instruction](https://windflow.online/use-cases/live-video-editing-from-a-webcam): Edit a live webcam or uploaded clip using text instructions. Compare SANA-Streaming and JoyAI-Video-Edit inputs, reference conditioning and responsiveness. - [Navigable world models for games: an API with WASD control](https://windflow.online/use-cases/navigable-world-models-for-games): Build navigable game scenes from a seed image with hosted world models. Compare movement, look, prompt conditioning and continuity limits on Windflow. - [Interactive AI storytelling with realtime video models](https://windflow.online/use-cases/interactive-ai-storytelling): Build a story users can direct while it plays. Choose scene or shot controls, keep narrative state in your app and budget realtime video sessions. Public product pages, model pages, guides, comparisons, use cases and field notes are available as Markdown: append `.md` to the URL (use `/index.md` for the homepage), or send `Accept: text/markdown`. ## Full public reference - [Full public text](https://windflow.online/llms-full.txt): public product facts, model specifications, guides, comparisons, use cases and field notes in one file; no account required. - [Sitemap](https://windflow.online/sitemap.xml): canonical HTML pages and source modification dates. ## Get access - Request access: https://windflow.online/auth?mode=request - Email: hello@windflow.online - Bring your own model: https://windflow.online/bring-a-model ## Pages - [Platform](https://windflow.online/platform): how sessions, control and metering work - [Models](https://windflow.online/models): the public model catalog - [Pricing](https://windflow.online/pricing): per-second rates and the FAQ - [Blog](https://windflow.online/blog): engineering field notes ([RSS](https://windflow.online/blog/rss)) - [About](https://windflow.online/about): who Windflow Labs is, and which other Windflows it is not - [FAQ](https://windflow.online/faq) and [Glossary](https://windflow.online/glossary): short answers and definitions - [X](https://x.com/windflowlabs) - [LinkedIn](https://www.linkedin.com/company/windflow-labs/) - [GitHub](https://github.com/windflow-labs) ## API documentation The full API reference, per-model command schemas and the Markdown docs index are served to Windflow accounts. Send `Authorization: Bearer ` to https://windflow.online/llms.txt for the docs index and to https://windflow.online/llms-full.txt for everything in one file.