Three sculptural silver film frames stand on a charcoal editing table; an amber correction line resolves into teal registration marks on the final frame. The headline reads Making Video Corrections Stick, with Framewright Studio Notes 04 above it.

This week, two finished films exposed a gap in my workflow. I liked their images and voices, but the lighthouse story still felt like a slideshow, while the street scene moved too slowly and its people and cars seemed to float above the road.

The production records already contained lessons about sustained movement. Writing those lessons down had not made the next film obey them. I needed a better connection between a previous correction and the checks performed before delivery.

That is the next improvement I want from Framewright: less repeated explanation when continuing or correcting a film.

What the first autonomous tests actually showed

I agreed the briefs for two simple films upfront: English Fola narration, horizontal and vertical versions, and complete publication packages. On the prepared Mac and Android-tablet setup, both runs delivered without requested intermediate approvals, downloads or rescue actions. Unsolicited owner messages remained in the record; this was not literally a conversation with zero input.

That is useful progress, but it measures a narrow kind of autonomy. It does not establish operation on an unfamiliar machine, a complete three-to-five-test programme, or lower production cost.

Same Street, Three Moods needed a faster overall pace and convincing surface contact. Its revised timeline runs twenty percent faster, with pitch-preserving audio retiming and matching subtitle times. The car wheels and planted feet were corrected against the actual road and pavement heights, then inspected in the encoded picture. The specific speed increase belongs to this film; it is not a new default for every narrator.

The lighthouse film needed a different repair. Camera drift, changing light and technically changing pixels had not supplied the activity I wanted. The revision introduced moving boats, waves, spray and reflections using retained short generated clips. Those clips repeat with transitions; I am not claiming a full minute of unique generated action.

I accepted both corrected films. The useful result is two completed examples and two concrete failure patterns, rather than proof of correction-free production.

Turn a preference into an observable check

The studio now keeps a small, source-backed set of video preferences. Each entry names where it applies, the correction behind it, what to inspect, and the exceptions.

For sustained scenes, I want meaningful subject or environment action across the full duration. A lively opening does not establish that the remaining minute works. For construction lines and captions, I want an encoded small-screen check over the actual background.

The Pantheon film supplied a particularly clear example. Its animated drawing looked too faint, especially vertically. The accepted repair strengthened the lines while protecting the soundtrack and later scenes. I kept the case's exact treatment in its recipe. The transferable lesson is to inspect readability at delivery size, not to mandate those colours or line widths everywhere.

I also distinguish preferences from functional defects. Feet hovering over a pavement require a contact repair. They do not justify inventing a new universal aesthetic rule. These records are working instructions for the next production, not an automatic judge of taste.

Preserve the conversation when changing the frame

Who's Interviewing Whom? exposed another format-specific problem. The original clip showed two participants. Enlarging only the active panel for a horizontal edit lost some of their shared context.

The correction kept the other participant in a small corner window, much like a video call. I used the existing captioned landscape film as the main input and preserved its timing and exact encoded audio packets. The accepted portrait file stayed unchanged.

The landscape picture still required re-encoding, so I do not describe its untouched areas as byte-identical. A useful revision record says precisely what survived: the same dialogue, timing and captions, with an added contextual view. The later owner-supplied publication link is recorded separately from the earlier vertical acceptance and the revised landscape's technical checks.

Carry uncertainty through the edit

The longer essay Before We Built It, We Imagined It needed the same discipline for facts. Its source map ties each episode to evidence. The historical connection between a rocket's lunar emblem and a particular film remains a hypothesis in the narration; an unverified film-set anecdote was left out.

Actual film excerpts and photographs sit alongside original graphic animation. A moving camera over a photograph remains an animated still, not continuous historical footage. Public reuse terms remain an open item in the production record, separate from my acceptance of the finished films and the supplied publication links.

The full narration also required an owner-completed download after the browser export stalled. Earlier tablet success did not make every later speech export autonomous.

Make the lesson usable outside this project

Alongside these productions, I shortened the current-work summary, separated completed history and connected corrections to pre-delivery checks. The two newest case records are still local, uncommitted additions as I prepare this issue. They are not a new release.

With Framewright, I keep the selected version, applicable preferences, revision request, protected elements and observations together in the existing project record. The editing tools still make the picture and sound.

Without Framewright, I would keep a versioned editor project, original sources, a short correction sheet, before-and-after exports and a delivery manifest in one ordinary folder. That parallel method is written here; it has not been independently reproduced by a reader.

For the next film, I want each relevant lesson to arrive as a concrete viewing task. Watch the sustained action. Check contact with the ground. Read the lines at phone size. Confirm what the revision preserved. A remembered correction becomes useful when someone actually performs that check.

Keep reading

Explore the writing archive →

Stay in touch

More essays and updates on ABVX Substack.

Read and subscribe on Substack →