Velorn: a desktop timeline editor that hands your edit to an MCP agent
AI-native video editing built around real creative timelines, generative workflows, and local agent control.
At a glance
- What is it?
- Velorn is an Electron video editor that pairs a multi-track timeline with ComfyUI generation and a large MCP tool surface for agents. The interesting part is the split: editing works standalone, generation does not.
- Who is it for?
- Adopt Velorn if you already run ComfyUI locally and want a timeline, an asset browser and an MCP surface in front of it, and if GPL-3.0 fits how you ship. Skip it if you have no ComfyUI instance, since every generation route depends on one, or if you need a mature script-to-scene pipeline.
- Can I use it commercially?
- Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
- Is it still maintained?
- Yes. The repository last received commits 5 days ago.
- What is it written in?
- Mainly JavaScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap Velorn is trying to fill between ComfyUI and a finished edit
ComfyUI produces clips. It does not hold a project. Once a graph has run, the outputs land in an output directory and the work of assembling them, timing them against music, captioning them and exporting a cut happens somewhere else. Velorn positions itself as that somewhere else. The README is explicit about the boundary: for generation, Velorn is not a replacement for ComfyUI, it is the production layer around it. Plan the work, send jobs to ComfyUI, collect the outputs, finish the edit.
The target user is narrow and identifiable. Someone making music videos from lyrics and timing, or UGC-style creator ads with a shot plan, or small-business ads. These are formats where the shot list is predictable, the generation is repetitive, and the assembly is mechanical. The README lists exactly those four Create workflows: Music Video Creation, UGC Creator, Business Ad Creator, and an experimental Short Film Creation that it describes as still very beta and possibly rough.
A second audience is the agent user. The tagline claims 100+ MCP tools, and the README shows a demo of an agent generating media, building the timeline and mixing audio live over MCP. That is a different proposition from a normal editor: the timeline becomes a programmable object rather than only a GUI surface. Note that the README states MCP editorial tools work without ComfyUI, while all current generation features require a locally running ComfyUI instance.
Two engines under one shell: Director Mode and the ComfyUI bridge
The architecture visible in the material has two generation paths feeding a shared editor.
The first is Director Mode, the engine behind Create. A Music Video project takes a song, lyric timing, characters and references, plus a director script, and expands them into keyframes, then video shots, then a timeline assembled from the generated shot assets. The README describes per-shot keyframe prompts, reference images, prompt copy and editing, image replacement, and shot reruns. That last set of controls matters more than the headline: a generation pipeline is only usable if you can rerun one shot without rebuilding the whole sequence.
The second path is the Generate workspace, which runs built-in local workflows, cloud and partner routes, and your own ComfyUI API graphs. There are two custom entry points, Custom Image and Custom Video, plus a raw API JSON import for people who prefer exporting from ComfyUI by hand. Custom video workflows can receive an injected keyframe image along with prompt, seed, width, height, FPS, duration and audio. That injection list is the actual contract between the editor and the graph, and it is what makes a third-party workflow usable inside a timeline rather than as a one-off render.
Bridging the two is Velorn Bridge, an optional component that lets a compatible graph open in ComfyUI be sent back into the correct Velorn panel. The Templates tab browses the official ComfyUI template catalog, described as 500+ templates with size and popularity info, and launches a chosen template into an embedded ComfyUI tab.
Installing it, and the config surface you actually touch
The README tells most users to download a packaged desktop app from the GitHub Releases page and to ignore the auto-generated source archives unless they intend to build from source. Release assets listed are Windows Installer, Windows Portable, Mac (Apple Silicon), Mac (Intel), Linux AppImage and Linux deb. The project is JavaScript on Electron, with React and the timeline editor among its topics, and it targets Windows, macOS and Linux.
There is no install command in the supplied material, and no documented configuration keys, so anything about a settings file or environment variables would be invention. What the README does describe is a workflow setup check that reports missing nodes, models, credentials and configuration. Treat that checker as the real onboarding step. The practical sequence it implies is: install the desktop build, have ComfyUI running locally, then open Generate, pick a workflow and read what the checker says is absent before queueing anything.
For custom graphs, the two supported routes are the Custom Image and Custom Video slots with Velorn endpoint nodes inside your ComfyUI graph, or importing an API JSON exported from ComfyUI. The endpoint nodes are how the editor knows where to inject the keyframe image, prompt, seed, dimensions, FPS, duration and audio. If your graph lacks them, the import path is the fallback.
The licence is GPL-3.0. That is a copyleft licence, and it governs the code you receive. If you plan to redistribute a modified Velorn, or to embed it in something you ship, the terms of that licence apply to your distribution. This is not legal advice; read the LICENSE file and, if the answer matters commercially, get proper counsel.
The timeline is a normal editor, and that is the point
Strip away the generation and Velorn still claims to be an editor. The README lists a project asset browser, a multi-track video and audio timeline, clip trimming, moving, snapping, overlap replacement behaviour and transitions, plus text, shape, title, solid-colour, adjustment-layer, keyframe and visual effect tools, an inspector, and proxy and cache tools. Captions and export are listed as working without ComfyUI, along with project management and the MCP editorial tools.
That separation is the most consequential design decision in the project and it deserves to be stated plainly. A user with no ComfyUI installation can install Velorn, import footage, cut it, caption it and export. They simply cannot generate anything. The README says so directly. Projects stay organised, with generated media, prompts, workflow outputs and timelines held inside a project rather than scattered across folders.
Proxy and cache tools are listed for a reason. Generated video at high resolution is heavy, and a multi-track timeline of it will not scrub smoothly on a laptop without proxies. The README mentions the tools but gives no detail on how proxies are generated, what codec they use, or whether the cache is per-project or global. That is a gap worth flagging: it is the kind of feature that is either fine or a daily irritation, and the supplied material does not let you tell which.
Where Velorn stops being the right tool
The clearest limitation is stated by the project itself: all current generation features require a locally running ComfyUI instance. Not a hosted one, not an optional one. If you are on a machine without a working ComfyUI, or you expected Velorn to be a self-contained AI editor, the generation half of the product does not exist for you. The cloud and partner routes named in the README, such as Nano Banana 2, GPT Image 2, Seedance and Kling, run through the same Generate workspace, but the README's own sentence about ComfyUI is unconditional, so treat cloud routes as something to verify on your install rather than assume.
The second limitation is maturity in the more ambitious workflows. Short Film Creation is labelled experimental and very beta, and the README says it may have rough edges. A script-to-scene coverage tool is a much harder problem than a shot-list ad, and the project is honest that it has not solved it yet.
The third is dependency fragility, which is inherent to the ComfyUI model rather than a Velorn defect. Custom nodes and models change. A workflow that ran last month can fail today because a node was renamed upstream. The setup checker catches missing nodes and models at configuration time, which softens this, but it cannot catch a node whose behaviour changed while keeping its name.
Finally, the release cadence is fast. Three releases in the six days before the last push, all v0.3.x. That is a project still moving, and pinning a version you have validated is more sensible than tracking latest.
How it differs from wiring ComfyUI to a general NLE
The obvious alternative is to keep ComfyUI for generation and use a conventional editor for assembly, moving files between them by hand or with a watched folder. DaVinci Resolve and Premiere Pro are the usual destinations, and both are far more mature as editors than a v0.3 project.
The difference in approach is where the project boundary sits. In the two-tool setup, the project lives in the NLE and the generator knows nothing about it. Prompts, seeds, workflow versions and the link between a timeline clip and the graph that produced it live in your notes, if they live anywhere. Rerunning a shot means finding the graph, remembering the inputs, exporting, and re-importing.
Velorn puts the project around both. Generated media, prompts and workflow outputs sit inside the project, and the timeline is assembled from the shot assets the pipeline produced. The custom video injection contract, with keyframe image, prompt, seed, dimensions, FPS, duration and audio passed into the graph, is what makes a rerun a single action instead of a manual round trip. The MCP surface extends the same idea to an agent, which can drive editorial operations directly rather than through the GUI.
That is a real structural difference, and it is also the source of the lock-in. A Resolve project is portable to other tools. A Velorn project is meaningful inside Velorn, and its generation half is meaningless without your ComfyUI install and the specific graphs you configured.
Maintenance cost and what to check before committing
Running Velorn means maintaining three things, not one. The app itself, which is on a rapid v0.3.x release line. Your ComfyUI installation, including the custom nodes and models each workflow depends on. And the workflows themselves, which are the part most likely to break silently when an upstream node changes.
The app is the easy part, since releases ship as installers and portables for all three platforms. The ComfyUI side is where the ongoing time goes. Every custom graph you import is a dependency you now track, and the README's setup check for missing nodes, models, credentials and configuration is the tool you will run most often. Budget for it as a recurring task, not a one-time setup.
Before adopting, verify four things on your own machine. That a release asset exists for your platform and architecture, since Mac builds are split between Apple Silicon and Intel. That your ComfyUI instance is running and reachable from the app. That the specific workflow you care about passes the dependency checker, including any partner or cloud route you intend to use. And that GPL-3.0 is compatible with how you plan to use or distribute anything built on top of it.
If those four check out and you are producing music videos or shot-list ads at volume, Velorn's bet on keeping the project, the prompts and the timeline in one place is worth the dependency burden. If you have no ComfyUI and no intention of running one, install it as a plain editor and ignore the Generate workspace entirely, because that is exactly what it becomes.
Editorial conclusion
Adopt Velorn if you already run ComfyUI locally and want a timeline, an asset browser and an MCP surface in front of it, and if GPL-3.0 fits how you ship. Skip it if you have no ComfyUI instance, since every generation route depends on one, or if you need a mature script-to-scene pipeline. Before installing, check the release asset for your platform, confirm your ComfyUI has the nodes and models your chosen workflow lists, and read the dependency checker output rather than assuming a workflow will run.
Community notes