Dear NUUG video-gruppe,
[drift: Looks like video.nuug.no is now out of service? I had rawdata
for the movie Fri programvare (2015) on the account ole(a)video.nuug.no,
but still can't login.]
Hence, I am reaching out with an idea to contribute to the restoration
and enhancement of Fri programvare (2015) by creating a new version of
the movie using generative AI. Given that the original visuals for the
2015 film are lost, I propose using the preserved audio recordings as a
foundation to recreate the film before 2025, with updated visuals
generated by AI. See https://ai.meta.com/research/movie-gen/
NUUG has demonstrated expertise in preserving and restoring film
archives, and I believe this project aligns well with NUUG’s mission to
make digital materials more accessible. By combining Meta’s generative
AI models like LLAMA and the Movie-Gen engine with a selection of
open-source models (similar to MyHeritage's Deep Nostalgia for images),
we could create dynamic moving visuals from still images. This approach
could make archiving classic films and documentaries more accessible to
the public, preserving them in open formats such as those supported by
Frikanalen.
Having developed open-source software centered on internet radio and
free audio mapping over the past decade, I am committed to creating
accessible tools that encourage exploration and preserve cultural
content. My initial experiments with tools like Autotekst from UiO have
shown promising results in processing audio from American radio, which
could similarly support audio-visual restoration.
Below is a Python script concept that leverages a generative language
model to create a basic movie script. This script shows how LLAMA could
dynamically generate dialogues for different scenes, enabling us to
bring Fri programvare (2025) to life through AI-generated scenes and
scripts.
The audio source material was archived, but the film was lost after 10
years at
https://archive.org/details/20151010-friprogramvarehttps://ia601302.us.archive.org/3/items/20151010-friprogramvare/20151010-fr…
I am attaching the audio transcription for the movie based on Dag
Langmyhr's interview in the audio of Fri programvare (2015) uploaded
and processed on https://autotekst.uio.no/ -- only Dag's transcript
from the first 1 minute was digitally preserved.
I envision combining generative AI scripts and archived URLs, like those
on archive.org and Frikanalen, to enhance historical and future film
content. With Meta’s Movie-Gen model and the LLAMA language model, I
believe it is possible to create an open-source implementation that
preserves important works while inviting new interpretations.
Please let me know if NUUG would be interested in collaborating on a
project to develop Fri programvare (2025) using these generative
techniques. I look forward to hearing your thoughts and exploring
potential next steps together, before I contact Meta's Llama Team.
Best regards,
Ole Aamot
ole(a)aamot.org
Aamot Research