Skip to main content Link Search Menu Expand Document (external link)

Retiming Text

Here’s a method for retiming text from normal to fast (“!!!”). It’s generic so applicable to any text, and yields a number of frames by which to adjust a gameplay segment timing, both for video and SGT retimes. Let’s call the run being retimed the subject run.

Rationale

Fast text replaces a sequence of normal textboxes with a single one saying “!!!”, and the task of closing this single textbox asap is part of the execution of an IL. This is substituted with the task of closing the first textbox asap in the subject run; the others are ignored, or equivalently assumed to be executed perfectly.

Calculation

We define three frames A, B, C to segment the text:

  • A: the first frame the first textbox appears
  • B: the first frame the first textbox starts fading out
  • C: the first frame the last textbox starts fading out

First, we have the raw retime, which is the subject converted to IL timing before adjusting for text. We take this, then

  • α: subtract the difference between the A to C time in the subject vs in an optimal run with fast-text
  • β: add the difference between the A to B time in the subject vs in an optimal run without fast-text

The α-adjustment swaps the whole text segment in the subject with that of an optimal IL, then the β-adjustment reinstates the time lost executing the first textbox.

The optimal timings we need to know are therefore:

  • A to C with fast-text. This should be always 8f (at 60Hz) and is hopefully level/language-independent.
  • A to B without fast-text. This depends on the textbox and language, but a good authority for JP (at 60Hz) is the 1:07 Any% TAS. E.g. this is 27f for JP Manta.

Once both adjustments have been calculated in frames, the final retime of the containing gameplay segment is done as follows:

  • for a video retime, the adjustment is deducted from the frame count, before it’s converted into a time by dividing by the frame-rate.
  • for an SGT retime, the IL is advanced backwards from the shine-get frame per the adjustment, then the time showing on the resulting frame is taken.

Example – JJ Manta IL

This example will use the times on the in-video RTA timer so that ppl can frame-advance the original YouTube video to understand what’s going on (press , and .). An accurate retime should be done (1) by counting frames in the video rather than using the timer, and (2) using times expressed in integer frames rather than decimal seconds until the final conversion (to avoid rounding errors).

Here are the important frames:

frame time
intro cs start 57:39.64
A (first text start) 57:50.82
B (first text end) 57:51.75
C (last text end) 58:03.44
shine-get cs start 59:13.78

and the calculation (note the optimal timings 8f and 27f mentioned earlier; the video is 30 fps exactly):

step calc result
raw retime 59:13.78 – 57:39.64 = 1:34.14
A to C 58:03.44 – 57:50.82 = 12.62
α-adjustment 12.62 – 8f = 12.35
A to B 57:51.75 – 57:50.82 = 0.93
β-adjustment 0.93 – 27f = 0.03
final retime 1:34.14 – 12.35 + 0.03 = 1:21.82

This run comes out to 1:21.80 using this technique with frame-counting.

Based off the TAS, here are reference times for optimal text with no fast text (A→B)

EP A→B
ap 28f
dsm 22f
g7 15f
q2 20f, 15f
q3 17f
q6 25f, 24f
q7 15f
p1 26f
p7 15f
r2 15f
r7 15f
b7 15f
s1 27f
s2 18f
s3 21f
s4 28f, 18f
s5 25f, 17f
s6 17f
s7 23f
n3 17f
n4 23f
n7 15f