Full Run Contract

The full production pipeline (-full / -run) produces a complete YouTube delivery set from one audio source and one image source.

Inputs

Place these in Output/ (or the configured SRC_DIR):

Exactly 1 source audio file: .flac, .wav, or .mp3 (archival *_RF64/*_BW64 companions are excluded from source selection). The run renames it to 1_source.<ext>, moving its companion family with it, and names every deliverable 1.*.
Source images are identified by orientation, not by filename:
- one landscape image — required, any size, upscaled to the 8K master
- one portrait image — optional, any size, used for the fitted shorts A second landscape or a second portrait is an error rather than a guess.
Image inputs, one of:
- exactly 1 source image: .png preferred, .jpg/.jpeg accepted, or
- direct 8K PNG inputs: Horizontal_8K.png for the main MP4, plus optional Vertical_8K.png for the short MP4

When both Horizontal_8K.png and Vertical_8K.png are present: the main MP4 renders from Horizontal_8K.png first, then the short MP4 renders directly from Vertical_8K.png. Direct Horizontal_8K.png inputs are used as-is for the main video but still derive companion image deliverables (4K PNG, NFT PNGs, 2K/3K PNGs, JPG exports).
When Vertical_8K.png is absent: the short MP4 uses or creates *_NFT8K.png, centers it in the portrait frame, and pads top and bottom with black.
Image and audio preparation run concurrently with bounded scheduling.

Image: 8K/4K PNG, NFT PNGs (8K/3K/2K), 3K/2K PNG, JPG exports (1MB/2MB/5MB/20MB targets)
Portrait stills: both short framings as artwork — *_Short_8K.png and *_Short_CenterCut_8K.png at full portrait resolution, each with a _1MB.jpg and _2MB.jpg export
Audio: internal-standard WAV (24-bit, 96 kHz, stereo RF64), ALAC M4A, 320 kbps MP3
Archival: *_RF64.flac, *_RF64.wav, *_BW64.wav (delivery-only; never reused as full-run source inputs). There is no *_BW64.flac — BW64 is a WAV-only container standard, so a FLAC counterpart could only ever duplicate the RF64 FLAC
Video: main MP4 (HEVC ladder, 7680x4320) and four portrait shorts (H264 ladder, 4320x7680, ALAC audio):
- _8K_Short.mp4 — image fitted inside the frame with black padding, capped at min(SHORT_MP4_CLIP_SECONDS, 58) s
- _8K_Short_CenterCut.mp4 — centre of the landscape 8K master scaled to cover and cropped, filling the frame with no padding
- _FullSong companions of both, rendered when the source audio is longer than the cap (a source within 0.01 s of the cap counts as being at it)

Input: two or more .mp3, .wav, or .flac tracks in SRC_DIR, sorted in natural numeric filename order, plus the same full-run image inputs
Each track is loudness-normalized to -12 LUFS, then joined with 2 s gaps into one RF64 album.wav
The album WAV then runs through the same full-run image/audio/video pipeline