OpenKatana

OpenKatana

OpenKatana is an independent, open-source editing engine your AI agent drives: your photos, footage and music in, a finished edit out; cuts, transitions, motion, typography and music in one shared project, all made with code. Inspired by Higgsfield's Katana, built with the Comfy Developer Platform, with local open-weights workflows and optional authorized API models. Version 0.1.3 #OpenSourceHiggsfield.

See it in motion:

Aura Farm Koji

Watch the anime edit with sound — the finale from the OpenKatana launch film. Source animation: “Koji” by Ammonite & Zean (CC BY); music made with Suno; edited with OpenKatana.

Motif

Watch or download MOTIF — 1080p, 60 fps, with sound

A 24-second product film with procedural 3D objects, choreographed cameras, animated displays, typography and an original soundtrack. OpenKatana offers code-authored WebGL rendering for 3D scenes, with a Blender option available too. This MOTIF export used the Blender option; OpenKatana assembles and finishes the film. No video-generation model was used. The GIF shows an eight-second preview at reduced resolution and frame rate; the MP4 preserves the original 1080p/60fps export and audio. Artwork and film credits.

Why OpenKatana is different

• Your agent edits, with real tools. Instead of one prompt-to-video model, your own AI agent cuts real footage with OpenKatana's editing tools: timeline, typography, camera moves, audio, 3D and effects. The result stays an editable project, not a flattened clip.
• One project, four ways in. ComfyUI Apps, Studio, the CLI and MCP open the same project files with revision checks, so you and your agent can take turns on the same edit.
• Match a reference edit. analyze_reference measures a reference at 60 fps (cuts versus flashes, picture cadence, the look of each shot, texture speed and the shape of each transition) and turns it into timeline settings.
• Compositing effects, written as code. Displacement (turbulent noise warps and map-driven heat haze, glass and water), distortion (wave warps, ripples, shockwaves, twirls, bulges, polar wraps, kaleidoscopes and mirrors), light (glows, lens flares, light rays and sweeps), time (echo trails, slit-scan time displacement, stepped frame rates and frame blending), blur and sharpen, stylize (RGB split, tiling, edge detection, emboss, mosaic and scatter) and generators (fractal noise, lightning, radio waves, cell patterns, gradients and grids). Every effect takes keyframes, eases and antialiasing, each as a few lines of timeline JSON.
• Open and local. Apache-2.0, runs on your machine, no account. Models are optional and installed separately.

New in 0.1.3: six effect packs for distortion, light, time, blur, stylize and generators

Watch all 37 effects or browse the effects gallery: Distort (Wave Warp, Ripple, Shockwave, Twirl, Bulge, Polar Coordinates, Kaleidoscope, Mirror), Light (Deep Glow, Lens Flare, Light Rays, Light Sweep), Time (Echo, Time Displacement, Posterize Time, frame blending), Blur & Sharpen (Gaussian, Directional, Radial and Camera Lens Blur, Unsharp Mask, Channel Blur, Tilt-Shift), Stylize (RGB Split, Motion Tile, Find Edges, Emboss, Mosaic, Scatter) and Generate (Fractal Noise, Advanced Lightning, Radio Waves, Cell Pattern, Gradient Ramp, 4-Color Gradient, Grid). They join Turbulent Displace and Displacement Map, and each one is a few lines of timeline JSON:

"fx": {"local_twirl": {"angle": [[0, 0, "ease"], [1, 240]], "radius": 0.45}}

Try any of them in ComfyUI with the OpenKatana · Preview effect node, or ask your agent to call preview_effect. The effects playbook has a recipe for each. Reproduce the reel with python examples/effects_reel.py (its effects are in examples/effects_reel.json; output in work/effects-reel).

Quick start

ComfyUI Manager: search OpenKatana, or run comfy node install openkatana with Comfy CLI configured for your ComfyUI installation. Then complete the editor setup and Apps installation. New Registry versions wait for Comfy's review; if Manager cannot resolve the latest version, use the source install below.

With your agent: copy and send the existing setup prompt:

Copy the setup prompt

Set up OpenKatana so you can create and edit videos for me with code.

1. Download the project with:
   git clone https://github.com/rk0neki/OpenKatana.git
2. Open the downloaded OpenKatana folder, read docs/SETUP.md,
   and run python scripts/setup.py.
3. Read the openkatana-video skill that setup installs for my first project.
4. If I use ComfyUI, install the OpenKatana nodes (docs/COMFY_APPS.md) and
   open the 00 Full workflow App first. Then check the required tools,
   open Studio, and tell me when it is ready.

Use the local CLI by default. Ask before downloading models, spending credits,
uploading files or publishing. Do not generate a video during setup.

Manual:

git clone https://github.com/rk0neki/OpenKatana.git
cd OpenKatana
python scripts/setup.py

OpenKatana runs on Windows, macOS and Linux. Python 3.11+ and Git are required. Setup prepares the Python environment, first project and agent skill. FFmpeg and a supported browser are separate installations (one command per platform); Blender and generative models are optional. Use python3 if that is your Python 3.11+ command and follow the next commands printed by setup. There is no OpenKatana login. Full setup guide · Optional MCP setup.

Use it in ComfyUI

OpenKatana on the Comfy Registry uses node ID openkatana and publisher rk0neki. See the Manager installation and editor setup steps. Registry review can delay version availability in Manager; the source installation above remains available.

Start with Apps → OpenKatana → 00 Full workflow. It puts every App in one graph: one project ID feeds Open Studio, your video, music and photo uploads, Make edit → Render to video, a 3D starter and any of the 75 tools. Run it from App mode and Studio opens in its own window.

OpenKatana includes seven saved Apps: Full workflow, plus Open Studio, All tools, Make a 3D scene, Background job, Video to edit and Preview an effect. The integration offers 88 nodes for its tools plus project, job and media helpers.

Connect a local model's image frames or video output to Import media → Make edit → Render to video, then open the same project in Studio to refine it.

After setup, ask your agent to follow the ComfyUI Apps installation guide. The Apps run locally; cloud and paid generation are disabled in that route. Authorized model calls are available separately through the CLI/agent workflow.

What can you make?

Turn a folder of photos into a story, cut footage to music, animate typography, or build a product film from a 3D scene. Your agent writes the edit and calls the tools; you set the direction and refine the result in Studio.

• Edits with rhythm: cuts, speed ramps, transitions, camera moves, beat analysis, audio mixing and sound effects.
• 3D films and morphs: code-authored WebGL scenes, materials, lights, camera paths, rigs and morphs, with an optional Blender rendering workflow.
• Motion graphics: animated lettering, multilingual text shaping, responsive layouts, masks and reusable effects.
• Finishing: frame previews, motion checks, HDR workflows and video export. Restoration and generation adapters use separately installed or authorized models.

OpenKatana offers 75 CLI/MCP tools (see them all). Your agent supplies the creative decisions; OpenKatana executes them and keeps an editable project on your machine.

How it works

1. Bring a brief and media. Describe the film, add footage, photos or music, or ask for an original scene made with code.
2. Let your agent author it. The agent plans shots, writes scene or animation code, and builds the timeline using OpenKatana's tools.
3. Render locally. Preview frames, render scenes and assemble the edit with typography, effects and sound.
4. Review and refine. Watch the result, give feedback, adjust it in Studio, and export when it is ready.

Studio, ComfyUI Apps, CLI and optional MCP work with the same project. You can move between agent instructions and hands-on review without starting over. The two AI-generated illustrations explain the product; the films above show actual rendered output.

How it's built

flowchart LR
  agent["Your AI agent<br/>(CLI or MCP)"] --> api
  comfy["ComfyUI Apps<br/>(Comfy SDK nodes)"] --> api
  studio["Studio<br/>(browser editor)"] --> api
  api["OpenKatana AgentAPI<br/>75 tools, strict JSON contracts"] --> project[("Project files<br/>timeline, media, revisions")]
  project --> engine["Render engine<br/>headless Chrome: Canvas 2D, WebGL,<br/>effect packs (web/fx_*.js)"]
  engine --> ffmpeg["FFmpeg<br/>MP4 export"]
  api -. optional .-> models["ComfyUI model workflows<br/>(local or authorized API)"]
  api -. optional .-> blender["Blender"]

Every entry point calls the same AgentAPI, so an agent, a ComfyUI App and Studio edit one project with revision checks. A timeline is plain JSON, validated against strict contracts (unknown fields are rejected). The engine draws each frame in headless Chrome, where the effect packs (distortion, light, time, blur, stylize and generators) register from web/fx_*.js, and FFmpeg encodes the result. The ComfyUI integration is built on the official Comfy SDK; model jobs run through ComfyUI.

Documentation and compatibility

• Setup · Studio · Agent guide · All tools
• Timeline · Editor tools · Effects · Effects playbook
• 3D · WebGL · Blender
• Compositions · HDR · Models
• Capabilities · Limits · Validation
• Contributing · Skill · Release notes

CI runs on every push and pull request, on Windows, macOS (Apple silicon) and Linux with Python 3.11 and 3.12: the regression suite, the bundled photo/video edit, an edit of the bundled LTX-2.5 clip (no ComfyUI or model run), the Studio browser workflow and a wheel installed outside the checkout. Check the completed runs for the commit you install. FFmpeg, Chromium, optional Blender and generative weights are separate installations. Studio is not a full graphical 3D scene editor. There is no hosting bill for the local app.

Experimental hosting and remote MCP

The hosted pilot prepares account workspaces, queued rendering and remote MCP. It is experimental and not deployed. Production identity, isolation, backups and model services need deployment-specific validation. See security and hosting costs.

FAQ

Do I need a GPU? No dedicated GPU is required for the standard editing and rendering workflow. Generation is optional; model and 3D workloads have their own requirements.

Does it cost anything? OpenKatana is free to use locally. Paid cloud models are optional; your chosen agent provider may charge for its service.

Which agents work? Claude Code and Codex are tested. Use an agent that can read local files and run commands.

How is it different from Higgsfield's Katana? It's open, local, edits your own media, and contains no Higgsfield code. Built for #OpenSourceHiggsfield.

Can I use it in ComfyUI without an agent? Yes, through the Apps.

Acknowledgments

Thanks to Comfy for the SDK and platform, Lightricks for LTX-2.5, the font authors, and NASA's media sources. Model and workflow credits · Third-party notices.

Independence

OpenKatana is inspired by Higgsfield's Katana and is not affiliated with, endorsed by, or sponsored by Higgsfield AI. It does not connect to Higgsfield's services. Those names identify the inspiration; their use does not imply endorsement or trademark rights. It contains no Higgsfield code, prompts, presets or assets. The project and its bundled dependencies have the licenses and source records linked below.

License

OpenKatana uses Apache-2.0. Bundled components retain their own licenses and notices; see NOTICE and third-party notices. External software, models and media have separate terms. The project license does not grant rights to them or guarantee legal clearance. See licensing.