Extensions/Reactor API: Realtime Video

ComfyUI Extension

Reactor API: Realtime Video

ComfyUI nodes that render timelines on Reactor real-time video models.

By reactor-team·Created 13 days ago·Updated a day ago· 5

reactor-team/comfyui-reactor-api

Nodes—

On cloudLocal install

Stars5

Updateda day ago

Readme

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

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

</div>

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.

<img src="example_outputs/reactor_realtime_window.jpg" width="560" alt="The Reactor Realtime window driving a LingBot World 2 world with the keyboard"> 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.

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.

<table> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Explore%20worlds%20-%20Real-time%20world%20driven%20with%20the%20keyboard%20%28LingBot%20World%202%29.json"><img src="example_outputs/reactor_realtime_world.webp" width="100%" alt="Real-time world driven with the keyboard"></a> Real-time world driven with the keyboard LingBot World 2 · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Explore%20worlds%20-%20Scripted%20camera%20moves%20%28LingBot%20World%202%29.json"><img src="example_outputs/reactor_camera_moves.webp" width="100%" alt="Scripted camera moves"></a> Scripted camera moves LingBot World 2 · 🎬 Render</td> </tr> </table>

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.

<table> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Real-time%20camera%20restyled%20from%20an%20image%20%28Vidu%20S2-Editing%29.json"><img src="example_outputs/reactor_realtime_camera_restyle.webp" width="100%" alt="Real-time camera restyled from an image"></a> Real-time camera restyled from an image Vidu S2-Editing · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Real-time%20camera%20dressed%20in%20an%20outfit%20from%20an%20image%20%28Vidu%20S2-Editing%29.json"><img src="example_outputs/reactor_realtime_camera_tryon.webp" width="100%" alt="Real-time camera dressed in an outfit from an image"></a> Real-time camera dressed in an outfit from an image Vidu S2-Editing · ⚡ Real time</td> </tr> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Real-time%20camera%20recast%20as%20a%20character%20from%20an%20image%20%28Vidu%20S2-Editing%29.json"><img src="example_outputs/reactor_realtime_camera_swap.webp" width="100%" alt="Real-time camera recast as a character from an image"></a> Real-time camera recast as a character from an image Vidu S2-Editing · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Real-time%20camera%20edited%20by%20a%20prompt%20%28X2%29.json"><img src="example_outputs/reactor_realtime_camera_edit.webp" width="100%" alt="Real-time camera edited by a prompt"></a> Real-time camera edited by a prompt X2 · ⚡ Real time</td> </tr> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Real-time%20video%20file%20edited%20by%20a%20prompt%20%28X2%29.json"><img src="example_outputs/reactor_realtime_video_file.webp" width="100%" alt="Real-time video file edited by a prompt"></a> Real-time video file edited by a prompt X2 · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Edit%20video%20-%20Restyle%2C%20re-dress%20and%20recast%20a%20video%20%28Vidu%20S2-Editing%29.json"><img src="example_outputs/reactor_vidu_edit.webp" width="100%" alt="Restyle, re-dress and recast a video"></a> Restyle, re-dress and recast a video Vidu S2-Editing · 🎬 Render</td> </tr> </table>

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. Stories across shots and cuts, one continuous scene steered by prompts, or a single shot that morphs, all from text.

<table> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20Real-time%20clips%20with%20sound%20from%20a%20prompt%20%28FastH3%29.json"><img src="example_outputs/reactor_fasth3_realtime.webp" width="100%" alt="Real-time clips with sound from a prompt"></a> Real-time clips with sound from a prompt FastH3 · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20Clips%20with%20sound%2C%20a%20shot%20then%20a%20cut%20%28FastH3%29.json"><img src="example_outputs/reactor_fasth3_clips.webp" width="100%" alt="Clips with sound, a shot then a cut"></a> Clips with sound, a shot then a cut FastH3 · 🎬 Render</td> </tr> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20Real-time%20clips%20of%20a%20person%20and%20a%20place%20from%20two%20pictures%20%28H3%20Reference%20Turbo%20Realtime%29.json"><img src="example_outputs/reactor_h3_reference_realtime.webp" width="100%" alt="Real-time clips of a person and a place from two pictures"></a> Real-time clips of a person and a place from two pictures H3 Reference Turbo Realtime · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20A%20person%20and%20a%20place%20from%20two%20pictures%20%28H3%20Reference%20Turbo%20Realtime%29.json"><img src="example_outputs/reactor_h3_reference.webp" width="100%" alt="A person and a place from two pictures"></a> A person and a place from two pictures H3 Reference Turbo Realtime · 🎬 Render</td> </tr> </table> <table> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20Real-time%20video%20from%20a%20prompt%20%28LongLive-2.0%29.json"><img src="example_outputs/reactor_realtime_prompt.webp" width="100%" alt="Real-time video from a prompt"></a> Real-time video from a prompt LongLive-2.0 · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20Nine%20segments%20of%20shots%20and%20cuts%20%28LongLive-2.0%29.json"><img src="example_outputs/reactor_nine_segments.webp" width="100%" alt="Nine segments of shots and cuts"></a> Nine segments of shots and cuts LongLive-2.0 · 🎬 Render</td> </tr> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20One%20scene%20steered%20by%20prompts%20%28Helios%29.json"><img src="example_outputs/reactor_helios.webp" width="100%" alt="One scene steered by prompts"></a> One scene steered by prompts Helios · 🎬 Render</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Generate%20video%20-%20A%20storm%20morphing%20around%20one%20shot%20%28Visko%20Orbis%20Stable%29.json"><img src="example_outputs/reactor_visko_orbis_stable.webp" width="100%" alt="A storm morphing around one shot"></a> A storm morphing around one shot Visko Orbis Stable · 🎬 Render</td> </tr> </table>

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.

<div align="center"> <table> <tr> <td colspan="2" width="100%" valign="top" align="center"><a href="example_workflows/Avatars%20-%20Real-time%20conversation%2C%20your%20voice%20in%20and%20video%20out%20%28Vidu%20S2-Avatar%29.json"><img src="example_outputs/reactor_avatar.webp" height="320" alt="Real-time conversation, your voice in and video out"></a> Real-time conversation, your voice in and video out Vidu S2-Avatar · ⚡ Real time</td> </tr> <tr> <td width="50%" valign="top" align="center"><a href="example_workflows/Avatars%20-%20Real-time%20takes%20of%20a%20photo%20speaking%20your%20script%20%28LTX%29.json"><img src="example_outputs/reactor_ltx_realtime.webp" width="100%" alt="Real-time takes of a photo speaking your script"></a> Real-time takes of a photo speaking your script LTX · ⚡ Real time</td> <td width="50%" valign="top" align="center"><a href="example_workflows/Avatars%20-%20A%20photo%20speaks%20a%20script%20in%20two%20takes%20%28LTX%29.json"><img src="example_outputs/reactor_ltx_takes.webp" width="100%" alt="A photo speaks a script in two takes"></a> A photo speaks a script in two takes LTX · 🎬 Render</td> </tr> </table> </div>

Models

Read the model's prompt guide before writing prompts: it applies to the prompts you write here, and each example's note links it.

Explore worlds

| Model | What it does | Takes | Prompt guide | | --- | --- | --- | --- | | LingBot World 2 | LingBot with more camera control | first image, camera moves | Prompt guide | | LingBot | A world you walk through | first image, camera moves | Prompt guide |

Edit video

| Model | What it does | Takes | Prompt guide | | --- | --- | --- | --- | | Vidu S2-Editing | Restyle, re-dress or recast from an image | video, image | Prompt guide | | X2 | Edit a video or camera from a prompt | video, optional image | Prompt guide | | Sana Streaming | Edit a video file from a prompt | video (Render only) | Prompt guide |

Generate video

| Model | What it does | Takes | Prompt guide | | --- | --- | --- | --- | | FastH3 | Short clips with sound, one per prompt | prompt, image on any segment | Prompt guide | | H3 Reference Turbo Realtime | Clips with sound, guided by up to nine pictures, which prompts call Picture 1, Picture 2, … in socket order | prompt, pictures | Prompt guide | | Helios | One continuous scene steered by prompts | prompt, image on any segment | Prompt guide | | Visko Orbis Stable / Dynamic | One unbroken shot that morphs, with sound | prompt, optional first image | Stable, Dynamic | | LongLive-2.0 | A story across shots and cuts, from text | prompt | Prompt guide |

Avatars

| Model | What it does | Takes | Prompt guide | | --- | --- | --- | --- | | Vidu S2-Avatar | Talk with a character made from a photo | first image, persona | Prompt guide | | LTX | A person from a photo speaks your script | first image, script | Prompt guide |

Not yet supported: HappyOyster.

Nodes

These are the nodes this plugin adds to ComfyUI, all under Reactor in the node menu.

| Node | What it does | | --- | --- | | Reactor Realtime | Runs a model in real time. Pick the model, queue it, and steer it in the window: edit the prompt and press Apply, drive the camera with W/A/S/D and the arrow keys, or talk to an avatar. Press Save to output the take as a video. | | Reactor <model> Segment | One segment of a scripted sequence, with that model's inputs. Chain segments through sequence. The first segment holds the model's settings. Since every input is a normal ComfyUI input, other nodes can write the prompts, make the images, or branch the story. | | Reactor Render | Renders a sequence and outputs a video. | | Reactor Sequence Join | Plays sequences for the same model one after another. | | Reactor Sequence Visualizer | Draws a sequence's segments and camera moves on a frame ruler. | | Reactor Vidu S2-Avatar Reference | Tags an image as an object, outfit or background for an avatar segment. | | Reactor Camera Capture / Microphone Capture | Pick this browser's camera or microphone for Reactor Realtime. |

Tips

Getting the same take every time? In Reactor Realtime, a fixed seed with unchanged inputs reuses the last take. Set it to randomize to record a new take each time.
Looking for a take? Reactor Realtime saves them to reactor/realtime in ComfyUI's output folder. Change it with the node's filename_prefix.
Talking to an avatar? Wear headphones, so the Vidu S2-Avatar character doesn't hear itself.
Want to run more than one workflow at once? ComfyUI runs one Reactor session at a time. Set MAX_CONCURRENT under [Sessions] in config.ini and restart to run more.

Contributing

Pull requests are welcome. Run the tests before opening one; they need no ComfyUI install and no Reactor API key:

uv run pytest -q tests
node --test "tests/web/*.test.mjs"

Sign off every commit (git commit -s): pull requests are checked for a DCO sign-off.

Licensing

This repository is Apache-2.0 licensed - see LICENSE.