Vlo 0.3 – An open source, extensible video editor and generator designed for AI compositing.

Hey all,

vlo is open source video editing and generation software. It is designed for interactivity between generative AI and the timeline, so you can inpaint anywhere on the timeline, extract frames and videos to use as references, use SAM2 for video masking and sam-audio for stem extraction etc. It comes with built-in workflows for Minimax H3, Qwen2.1, Krea2, LTX2.5 and more.

You can either use the built-in generation panel or open ComfyUI via the app, whichever you prefer (for built-in workflows though, I recommend the generation panel). It can use an existing ComfyUI install, a remote instance, or it can manage the install for you; ComfyUI is the primary AI engine, how much you directly interact with it is up to you.

I believe what makes this app different is its emphasis on control and correction over automation, so particular effort has been spent building a frame-accurate render engine (which is nontrivial for web video), and on smoothing out irregularities you get from things such discrete model strides and such (i.e. aspect ratios are not arbitrary in video and image gen models, so if you want seamless passing of data with the timeline you have to be tactical about how you stretch, squash or crop). It permits patch-based inpainting without ever having to reencode areas outside the patch until final export, so you could inpaint a dozen objects without degrading unedited pixels at all.

It also has an SDK for you to build your own extensions and shader-based special effects. I hope to upload a couple of example extensions soon.

Installation instructions here:

https://github.com/PxTicks/vlo#install