Imported from GitHub: joeyparrish/kinetoscope · commit d16046b · license MIT

Description

Kinetoscope: A Sega Genesis Video Player

README

Sega Video Encoder

Encode videos into a format appropriate for streaming to a Sega Genesis.

Prerequisites

The encoder requires:

• Python 3
• the PIL Python module
• a copy of ffmpeg with PNG support

On Ubuntu, you can install these with:

sudo apt install python3 python3-pil ffmpeg

Instructions

To make a 10-second clip, from time 0 to time 10, run:

./encode_sega_video.py \
  -i /path/to/never-gonna-give-you-up.mp4 \
  -s 0 -e 10 \
  -o res/video_data.segavideo \
  --generate-resource-file

Adjust the input file (-i), start time (-s), and end time (-e) however you like, but you can only fit about 13.6 seconds on a 4MB ROM with enough room left for the player.

The video output file (-o) contains the actual video stream. If you generated an SGDK resource file (--generate-resource-file), it will have the same name as the output, but end in .res. SGDK will generate a header file next to it that ends in .h, which you will reference in your code to get the address of the video in ROM.

For a complete example of embedded video in a ROM, see ../software/embed-video-in-rom/.

To prepare video for a streaming server (server/ folder), omit the --generate-resource-file flag, as well as the start and end (-s and -e) flags, and add the --compressed flag.

Other Settings

You can see a full list of settings with ./encode_sega_video.py --help. Here are a few that require additional notes:

• --fps: The maximum video frame rate we have managed to sustain is 10fps. Above that, we drop frames, which risks breaking the careful synchronization of chunked streaming between the Sega's M68k CPU and the streaming coprocessor.

• --sample-rate: The audio sample rate is technically not fixed in the video file format, but the only performant audio driver that we have gotten to work so far is SGDK's XGM2 driver, which has a fixed sample rate of 13312 Hz. If you use a different rate, the audio will play at the wrong speed. A new Z80 audio driver will be needed in the player to lift this restriction.

• --chunk-length: The default chunk length is tuned to make chunks at the default frame rate and sample rate fit inside the SRAM buffer of our special hardware. If the chunk size exceeds 1MB, the streaming hardware in hardware/ and the emulation of it in emulator-patches/ will not work.

• --scene-detection-threshold: The sensitivity of the scene-change detection, which is used to optimize color quantization by choosing a stable palette per scene. The value is a ratio of changed pixels that constitutes a new scene, from 0-1. A threshold of 0 disables scene detection and creates an independent palette for each frame.

• --dithering: The dithering algorithm, which controls how the limited palette of 15 colors per frame are used to represent the full color original image. The default is "none". For a full list of options, see https://ffmpeg.org/ffmpeg-filters.html#paletteuse