Skip to content

Make captions again once a project has some - #187

Merged
bpacholek merged 1 commit into
mainfrom
181-captions-again
Nov 1, 2026
Merged

bpacholek merged 1 commit into
mainfrom
181-captions-again

Conversation

@bpacholek

Copy link
Copy Markdown
Contributor

Closes #181

Once a project had captions, Auto captions couldn't be run again: the phone's Captions tool opened the Captions screen, which had no way back, and the tablet offered Auto captions only while there were none. So clips added later couldn't get captions, and captions couldn't be made again with another model or language.

Phone. The Captions screen's model line ("Whisper Small · Polish ⌄") is now a button to the Auto captions setup. The CaptionsEdit board has no control for this, so this is the difference from the design: the line was already there and says what the captions were made with, so it leads to where that is changed (ScreenHeader takes an optional onSubtitle, with a chevron and a TalkBack label). Generating returns to the Captions screen it came from. Above Generate, the setup says how many captions it replaces: "Replaces the 6 captions made from these clips, edits included".

Tablet. The captions inspector has the Style and Transcribe tabs of TabletCaptions--transcribe-tab: the Whisper models with their real sizes and state, the language, Translate to English, and "Regenerate with Base" ("Generate with Base" when there are no captions), with progress and Cancel while it runs. It hears the clips with captions again, plus the main clips, so clips added since get captions. Difference from the board: the language and Translate sit one under the other, because side by side the language was cut off ("Auto-det…", "Autom…" in Polish), and a model's size goes under its name when both don't fit.

Running again (CreateCaptions, core:model), one undo step:

  • the captions made before from the same clips are replaced, with their translations;
  • the new ones keep the look of the captions (style, position, size), instead of going back to Bold pop;
  • they go on the layer the old ones were on when nothing there is in the way, even when other clips share it; layers the old ones leave empty go;
  • the captions as heard are shown again (a translation may have been on).

Tests:

  • CaptionCommandsTest.runningAgainKeepsTheLookAndLayerAndCaptionsNewClips
  • CaptionsSetupScreenTest.generatingAgainReplacesTheCaptions
  • CaptionsEditScreenTest.captionsAsDesigned taps the model line
  • CaptionsPanelsTest.tabletTranscribe and tabletTranscribeInPolish (new screenshots); the other captions screenshots were recorded again for the chevron and the tabs

Checked on the emulator with a real project (captions from a talk in Polish made with Whisper Small, Karaoke, on a layer that also holds text clips):

  • phone: Captions, then the model line, opens Auto captions with "Replaces the 6 captions made from these clips, edits included"; Generate with Base made 11 captions (the main clips include street clips with people talking), kept Karaoke and its size, and came back to the Captions screen; the new captions went on a new top layer because the text clips on the old one were in the way;
  • tablet: the Transcribe tab as above, Regenerate with Base, the same result;
  • one undo after each brought the project back exactly (layers, links, transcripts compared with a copy taken before).

🤖 Generated with Claude Code

The phone's Captions screen had no way back to Auto captions, and the
tablet offered it only while there were no captions. The phone's model
line now opens the setup, which says how many captions it replaces, and
the tablet inspector has the Transcribe tab of the design. Running again
keeps the captions' look and layer, drops their translations and
captions the clips added since.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@bpacholek
bpacholek merged commit 1349dfc into main Nov 1, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Auto captions can't be run again once a project has captions

1 participant