Skip to content

Field report 1 3 - #62

Merged
dkackman merged 3 commits into
masterfrom
field-report-1-3
Sep 11, 2026
Merged

Field report 1 3#62
dkackman merged 3 commits into
masterfrom
field-report-1-3

Conversation

@dkackman

Copy link
Copy Markdown
Owner

No description provided.

dkackman and others added 3 commits September 10, 2026 22:54
assemble-and-score and dissolve-between-shots ran every shot through
stabilize_video with smooth 0 under a description that said it only
re-encoded them. smooth 0 is a full first-frame lock: on a shot with a
deliberate camera move it fights the move, crops and rescales every frame,
and nothing downstream notices because duration, size and sample rate all
survive (measured 13-20 dB PSNR against the source vs 45 dB for a plain
cut). The hold steps are gone; the shots go straight into the join, as
one 'shots' list variable so a diptych is two entries and a reel is however
many the cut needs. concat_videos and dissolve_videos now refuse shots of
different sizes by name, which is the check the hold steps were imagined
to be. The catalog shape rule counts a variable-fed or asset-fed cut as a
sequence that needs input media, so the templates keep their listing.

audio_bleed_ms and seam_fade_ms are variables in dialogue-short,
music-video and assemble-and-score, so a seam is re-tuned with an argument
rather than an inline copy of the workflow; dialogue-short defaults to
1800 ms, where a five-shot H3 cut measured best (700 ms left a 44 dB hole
at the worst seam, 1800 brought it to 32, 2500 gained almost nothing).

validate_workflow warns when concat_videos is given a crossfade_ms with
nothing trimmed - the crossfade is drawn from the trimmed material, so at
a hard cut it reads as active and does nothing. The concat_videos and
stabilize_video docs say all of this.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@dkackman
dkackman merged commit d3f808c into master Sep 11, 2026
4 of 5 checks passed
@dkackman
dkackman deleted the field-report-1-3 branch September 11, 2026 04:03
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.

1 participant