mstudio
A desktop video studio where creators can begin with an incomplete idea and discover what they want through scripts, storyboards, and actual footage.
THE QUESTION
How can an AI creative tool help people move forward before they know exactly what they want, while preserving the parts they already like?
DESIGN & PRACTICE
- Made scripts, the canvas, and the timeline editable working objects so that intent can develop after people see a result.
- Organized conversation and direct editing around specific objects, letting creators choose when to delegate and when to adjust the work themselves.
- Separated specialist judgments, narrowed the scope of local changes, and removed a separate review step that was not reducing the work of checking results.
MSTUDIO / PRODUCT DESIGN NOTES · OCT 2026
How can a product help before people know exactly what they want?
A storyboard can change the composition someone wants. Actual footage can reveal a problem with action or pacing. In my own production work, many requirements emerged only after I saw a result. mstudio is organized around that process: express an idea, see the work, make a judgment, and keep revising.
Let intermediate work help people discover intent
Scripts, storyboards, and the timeline make story, composition, and timing available for judgment. I kept these objects editable so people can decide what comes next after seeing the work. They can revise a stage or finish only the part they currently need.
Choice: keep useful creative surfaces while removing technical nodes and separate approval flows that users would have to manage.
Move between delegation and direct editing around the same work
A creator can reference a shot and describe a change, or adjust a clip directly on the timeline. Both act on the same project, so the person does not need to transfer a chat suggestion into the editor again. Agents handle execution while the creator retains judgment over the image and its timing.
Choice: offer one working context and call specialists as needed, without requiring every task to pass through the entire production process.
Preserve the parts people already like
A task to change shot 01 once expanded into a whole-film review. Editing a generation task also overwrote a shot draft. These incidents made the target and reach of a change explicit product requirements. People should be able to predict what their instruction will change and what it will preserve.
Choice: treat a task description, a reusable shot draft, and a new generation as separate actions. Reading more context must not silently expand the edit.
Keep specialist coordination inside the product
Deciding how a shot communicates an action differs from deciding how to draw a particular frame. I separated shot direction from storyboard artwork so each role could address its own judgment. Creators still express goals around scripts, shots, and clips; the system takes on coordination between roles.
Choice: add specialization where a distinct judgment is needed; disable a separate review role when it does not reduce my own checking work.
Evaluate automation by the work that remains
A continuation test on September 27 produced a 23-second film, but required interface confirmations, external frame inspection, and corrected cut points. A completed first edit did not mean the actions had been preserved. Those incidents help me examine how much work a person still has to do after the system finishes.
Choice: distinguish saved changes, a successful export, and content quality. A finished task is not evidence of a satisfactory film.