main @ 70b62bd — These are the reels every workflow change is compared against.
Render time
44 s
CPU
735 s
Avg cores
16.7
Speed
65.2 fps
Peak memory
4.96 GB
Golden reels
01_fiction_platforms
Reference render
main
Length
25.2 s
Render time
28 s
CPU
141 s
Cores used
5.1
File size
16.4 MB
02_retcon
Reference render
main
Length
35.2 s
Render time
39 s
CPU
267 s
Cores used
6.8
File size
21.0 MB
09_burn_bridges
Reference render
main
Length
28.8 s
Render time
33 s
CPU
190 s
Cores used
5.8
File size
12.1 MB
10_infosys
Reference render
main
Length
25.4 s
Render time
16 s
CPU
137 s
Cores used
8.5
File size
16.5 MB
How to approve
Watch the reels above. Happy? Approve and merge the pull request — the change is on main.
Then run reelflow baseline on main so these become the reels the next change is compared to.
Not happy? Leave a comment on the pull request; nothing reaches main until it's merged.
How reels are judged. Timing is checked exactly, from every word's position on the reel's timeline.
The picture score is SSIM with film grain blurred out — "looks the same" is ≥ 0.99; the same code rendered twice scores 0.9994,
a visible grade nudge 0.9886. "Fine detail" is plain SSIM and only informational: grain re-encoded with different encoder
settings drops it to ~0.985 with no visible difference. ffmpeg's text overlay can also land a cut one frame early or late
from run to run (it did before reelflow too), so a couple of frames at cuts may differ in any comparison.