Six finished films: HyperFrames workflows, public source and render provenance

[cryptokali007]

Hi HyperFrames folks! We're sharing a collection of six finished films, together with their editable source and release pages. The repositories include editable compositions, project-specific media, dependency records, upstream notices and production provenance.

Browse the film portfolio and previews. Each film has downloadable viewing copies and its own source repository.

This is a collection of related workflows, not six examples of an identical pipeline. The Shape of Remembering, The Space Between and This Film Has a Memory were produced locally with HyperFrames. The Memory Between was locally authored, rendered on Kaggle GPUs and finished locally. The two Kelpie projects used Kaggle-hosted Astra/xhigh for source authoring, with GPU picture rendering recorded as a separate stage.

Watch and inspect the source

Film	Length / frame rate	Source	Release
Kelpie In Motion	12 seconds / 60 fps	Repository	v1.0.0
Kelpie: The Shape of Patience	5 minutes / 60 fps	Repository	v1.0.0
The Memory Between	5 minutes / 60 fps	Repository	v1.0.0
The Shape of Remembering	5 minutes / 30 fps	Repository	v1.0.0
The Space Between	72 seconds / 30 fps	Repository	v1.0.0
This Film Has a Memory	5 minutes / 30 fps	Repository	v1.0.0

Three engineering details worth discussing

1. One frozen composition, two picture ranges. Kelpie: The Shape of Patience has 18,000 frames at 60 fps. Two concurrent Kaggle notebooks rendered 0–150 seconds and 150–300 seconds from one frozen source hash, with a T4 pair assigned to each notebook. Hosted Astra/xhigh authored the 20-shot story and original procedural horse module; the notebooks executed the picture renders. Keeping those roles separate makes the production history easier to inspect.

2. Audio analysis and local finishing were explicit stages. The Shape of Patience used a Kaggle GPU asset job for its original score and water maps, local CUDA Kokoro Heart narration, and actual STFT/RMS audio-band analysis. Local joining, muxing, decode checks and an NVENC viewing copy followed the picture renders. Its caption timing is estimated—a useful limit to retain in the documentation rather than hide behind a general reproducibility claim.

3. A publication transcode is not the original production pipeline. For the current publication of The Memory Between, the full accepted 4K master was copied via authorized SSH to Colab G4 with an RTX PRO 6000 Blackwell GPU. CUDA resizing and NVENC produced the full 1080p60, 18,000-frame derivative, with the original AAC copied and the separate SRT preserved. Full-decode and hash checks were part of that pass. Colab was used for this publication derivative, not the film's original authoring or rendering. Its public lockfile was also newly generated from original direct pins, not recovered from the historical production environment.

Scope and credits

Credit to HyperFrames, Three.js and GSAP for the composition tools behind the deterministic HTML/scene work, and to the specific tooling used in the stages above: Kaggle GPU notebooks, Kokoro Heart, CUDA, NVENC and Colab. The repositories retain their original software, framework, font and brand notices; this post grants no blanket license.

Not everything in the collection is procedural: The Memory Between includes supplied closing media alongside its locally authored composition, original score and narration. Kelpie In Motion is conceptual marketing imagery, not a protocol implementation or live financial UI. This Film Has a Memory uses curated evidence from preproduction MemWal recalls and a separate checkpoint; its protocol illustrations are conceptual.

The repositories have not all been independently reproduced on a clean machine. We'd welcome technical feedback on the source organization, deterministic timeline work, split-render provenance and caption documentation. Code and documentation feedback are just as welcome as render tests. Thanks for taking a look!