Sunbelt Computer Software

Reactor for ComfyUI — world models in your workflow

Real-time video models in ComfyUI. Steer them live, or script every shot.

🌐 Reactor · 📚 Docs · 🛠️ Client SDKs · 📖 Cookbook

A ComfyUI custom node pack for Reactor's real-time video models. Play a model live in a window you steer as it generates, or script a sequence of segments and render it to a video, all from a normal ComfyUI workflow.

_{LingBot World 2 playing in the Reactor Realtime window, steered with W/A/S/D and the arrow keys.}

Getting started

1. Install. Clone this repo into ComfyUI/custom_nodes/ and install its requirements into ComfyUI's Python:

cd ComfyUI/custom_nodes
git clone https://github.com/reactor-team/comfyui-reactor-api
pip install -r comfyui-reactor-api/requirements.txt

2. Add your API key. Copy config.ini.example to config.ini and set your Reactor API key, or set REACTOR_API_KEY in ComfyUI's environment (it takes priority).

3. Copy the example inputs. Copy example_inputs/ into ComfyUI's input folder so the examples find their images and sounds.

Examples

Every example is in ComfyUI's template browser under comfyui-reactor-api, or drag a preview onto the canvas to load its workflow. Edit previews show the source on the left and the edit on the right.

Each example runs in one of two ways:

⚡ Real time — the Reactor Realtime node opens a window where the model plays as it generates. Edit the prompt, drive the camera with the keyboard, feed it your webcam, or talk to an avatar.
🎬 Render — chain segment nodes into a sequence and Reactor Render turns it into a video. Other nodes can write the prompts, make the images, or branch the story.

Explore worlds

Walk through a world made from a single image, steering the camera live with the keyboard or scripting every move ahead of time.

_{Real-time world driven with the keyboard
LingBot World 2 · ⚡ Real time}

_{Scripted camera moves
LingBot World 2 · 🎬 Render}

Edit video

Restyle, re-dress or recast your camera or a video file from a single image, or edit it with a prompt. Source on the left, result on the right.

_{Real-time camera restyled from an image Vidu S2-Editing · ⚡ Real time}	_{Real-time camera dressed in an outfit from an image Vidu S2-Editing · ⚡ Real time}
_{Real-time camera recast as a character from an image Vidu S2-Editing · ⚡ Real time}	_{Real-time camera edited by a prompt X2 · ⚡ Real time}
_{Real-time video file edited by a prompt X2 · ⚡ Real time}	_{Restyle, re-dress and recast a video Vidu S2-Editing · 🎬 Render}

Generate video

Short clips with sound from a prompt with FastH3, or from pictures of a person and a place with H3 Reference Turbo Realtime. FastH3-Omni adds audio input and first-to-last frame to reference images. Stories across shots and cuts, one continuous scene steered by prompts, or a single shot that morphs, all from text.

_{Real-time clips with sound from a prompt FastH3 · ⚡ Real time}	_{Clips with sound, a shot then a cut FastH3 · 🎬 Render}
_{Real-time clips of a person and a place from two pictures H3 Reference Turbo Realtime · ⚡ Real time}	_{A person and a place from two pictures H3 Reference Turbo Realtime · 🎬 Render}
_{Real-time talking video with audio input FastH3-Omni · ⚡ Real time}	_{Nine reference images, one scene FastH3-Omni · 🎬 Render}

_{Real-time video from a prompt LongLive-2.0 · ⚡ Real time}	_{Nine segments of shots and cuts LongLive-2.0 · 🎬 Render}
_{One scene steered by prompts Helios · 🎬 Render}	_{A storm morphing around one shot Visko Orbis Stable · 🎬 Render}

Avatars

Talk with a character made from a photo: your voice goes in, video comes out. Or have a photo speak your script, one take at a time.

_{Real-time conversation, your voice in and video out Vidu S2-Avatar · ⚡ Real time}
_{Real-time takes of a photo speaking your script LTX · ⚡ Real time}	_{A photo speaks a script in two takes LTX · 🎬 Render}