Python script to time semi-automatically the loading screens of run submitted for Windblown.

RTA (Real Time Attack)/LRT (Load Removed Timer)

Definition

RTA: Wall-clock timer (user) that contains the entire run, without interruption.

LRT: Timer that does not include loading screens, but includes Real Time Cinematic and main/pause/inventory menu (occur in-game engine).

Loading time

Loads are delimited by the "purple frames": vertical gradient of dark-purple to purple. When the frame does not exist in the footage, the latter frame have to be used when entering the load, the earlier frame have to be used when leaving the load: no in-engine element should be visible. For a single run, the loads to remove are:

• from biome 1 to biome 2
• from biome 2 to biome 3
• from biome 3 to biome 4

Purple frames

The exact frame

Don't bother further if you find this:

A frame outside the loading screen

Not acceptable:

A frame inside the loading screen

Acceptable:

Windblown-s-RTA2LRT

Windblown-s-RTA2LRT is an open source python script that comes with a GUI, allowing players to semi-automatically seek for the "purple frames". Once a video is fully processed, it returns the load duration (quantized in frame duration), the LRT is defined as RTA - Load duration. Runners are encouraged to re-time the runs themself and submit both the RTA and LRT. When they do so, they must send the output files produced by the tool in the general-chat channel of the Community Discord server and ping a moderator. Otherwise, the moderation will do it themself.

User manual

Installation

Requirements are fully detailed on the Github page: https://github.com/not-ngl/Windblown-s-RTA2LRT.

A video example installation is also provided on a fresh Windows 10.

Load the video

Either local file or link.

Local file recommendation

Bigger resolution makes the program slower. It is recommended to resize the video to 360p, typically using ffmpeg or your favorite editing software.

Download file

Make sure to be connected to your Google Account. If you have troubles because of "Chrome cookie database", check: https://github.com/yt-dlp/yt-dlp/issues/7271. The file is downloaded in the lowest available resolution at maximum 30fps to limit bandwidth.

Spectrogram

Once the audio is fully processed, the spectrogram is displayed in a timeline. If you were not talking during loading screens, they should appear quite, making it easier to spot with the eyes. Click on the silent regions: it automatically creates an anchor. It also displays the frame of the time selected: ensure it is a loading screen.

Expansion

Run the parallel expansion: it seeks forward and backward for boundaries of load screen. It is important that your guesses are well within the loads (no fade-in/out). If parallel mode is problematic, it can be disabled.

Manual refinement

Adjust the boundaries of identified regions using

• h (-5 frames),
• j (-1 frame),
• k (+1 frame),
• l (+5 frames)

Use Tab/Shift+Tab to go to the next/previous instant to adjust. Once you are done, use the export button to produce the output files in a dedicated directory (named following the timestamp when you exported YYYY-MM-DD-HH-MM-SS: provide all the files from there).

User Guide Video

Check the video below for:

• full installation on Windows
• full installation on Linux
• usage example