Workflow

Local or Cloud, Per Clip: The Model-Selection UX Problem No Editor Has Solved

23 September 2026 · RenderBob team

The right model changes inside one timeline: a local preview for a test, a cloud model for the hero shot. A global toggle forces the wrong cost onto every clip.

Any editor that can run a local model and a cloud model has to let you choose, at the speed of a cut. Most of them hide that choice in preferences.

A global toggle is the wrong grain

The right model changes inside one sequence. A quick test of whether an idea works at all fits a local model on Apple Silicon: fast, free, private, good enough to judge a direction. The client-facing hero shot in the same timeline may need a frontier cloud model, and be worth the cost and the wait for that one clip. A single preference either spends cloud money on every throwaway or runs the hero on the weaker model. Flipping the preference back and forth is the settings-menu break that pulls you out of the edit.

Drift makes the switch urgent

Reference fidelity fails differently on different models, local and cloud. When a cloud generation ignores the reference, the next move is the same brief on another model, without re-entering the prompt, re-uploading the image, or starting a separate app.

Put the choice on the clip

Model choice belongs on the generation panel for that clip. One click, mid-session, and the prompt and reference stay. Defaults can follow the job and still be overridable: a preview starts local, a clip marked final starts on a cloud model. At the moment of the click, show expected time and, for cloud, expected cost. If the local pass misses, retrying on cloud, or the reverse, should be as short as the regenerate habit in the previous article.

The unsolved part is making that choice feel like editing. RenderBob's timeline keeps it on the clip: local Apple Silicon inference or a cloud model, per generation, with the result on its own lane next to the preview.