platoseedplatoseed
type text, dream video
Linum builds generative video models.
Linum is a small research-led startup that trains text-to-video models from scratch and shares model weights and field notes. It positions itself as a two-person lab focused on generative video research and real-world lessons, with Apache 2.0 licensed models released publicly.
Linum develops and releases text-to-video models trained from scratch, including 2B-parameter variants (Linum v2) with 360p and 720p outputs. The site shows downloadable model weights and provides field notes documenting research, releases, and practical lessons. They publish blog-style write-ups and provide access to source code and model weights under an open license (Apache 2.0).
Who it’s for: researchers, academia, and small teams interested in text-to-video generation and open-source model weights
publication of model weights and technical field notes; ongoing research outputs and blog-style posts
Sahil graduated from Stanford University (BS, MS '19), where he co-wrote the graduate level course on LLMs and pursued cognitive science research at the intersection of AI and linguistics. He is a Bay Area native who loves movies, the Dubs, and burritos. https://thesahilchopra.com/
Manu graduated early from UC Berkeley (BA '21). Prior to founding Linum, Manu pursued research at the intersection of AI, bioengineering, and healthcare (e.g., applying offline reinforcement learning to learn optimal strategies for anesthesia administration). He is a Bay Area native who loves movies, the Dubs, and ice cream.
Formerly “StoryTime” · why startups rename →
interactive video models
Luma Labs helps people turn ideas into stunning visuals, powered by innovative research on image and video foundation models.
