Mago Studio restyles video on your GPU, with no account for local rendering

A free video editor that uses your own GPU comes with a wonderfully physical footnote: the models need somewhere to live. Mago Studio’s Windows setup calls for about 63 GB of free storage before local rendering is ready. The subscription bill can disappear while a substantial part of the drive develops a new tenant.

The desktop project’s README offers two rendering paths. Local work goes through ComfyUI on your hardware; cloud models require a Mago account and payment. The welcome screen says “Press Skip” for offline work. That makes the account optional for the local route, rather than an unavoidable ticket to an application installed on your own computer.

Its Transform workflow works on footage already shot: selected frame ranges, masks, relighting, restyling and character replacement. Those controls matter because changing one part of one shot is a different job from handing an entire clip to a generator and hoping it remembers the furniture. The editor keeps shots and versions in a project, with review tools around the generated results.

Windows gets the bundled setup path. The wizard checks the GPU and disk space, requests acceptance of the model licenses, and downloads ComfyUI, the required nodes and model files. The listed minimum is an NVIDIA GPU with 8 GB of VRAM. Mac and Linux users point the app at their own ComfyUI installation, so “runs on three platforms” hides a very different first afternoon on each one.

The project recommends its portable ComfyUI bundle because existing installations can have mismatched nodes or Python versions. Anyone already maintaining a carefully balanced workflow should read that as useful advance warning. The editor being friendly does not make every dependency in an existing model environment equally sociable (Python has opinions).

Local video rendering is free in the sense that Mago charges no cloud-render fee for it. GPU time, power and storage still belong to the person running the machine. The documentation offers no universal local completion time, so a minimum-memory check cannot be translated into a promise about how quickly a whole sequence will finish. One representative shot is a better audition than a promotional montage.

The cloud path also changes the privacy calculation. Mago says a cloud render uploads the shot and the material needed for it, including source video, reference images and masks, while the rest of the project remains local. Keeping a project on disk therefore does not by itself establish that no footage leaves the computer; the render destination is the choice to inspect.

The app’s AGPL license and downloaded model licenses remain separate. Installing an open-source interface does not rewrite the terms attached to its models. For someone experimenting with their own footage, the appeal is control over where a render runs and what it costs to repeat. Whether the local machine delivers an acceptable result at an acceptable speed remains the trial nobody else’s screenshot can perform.

The render queue moved into the house.