AI Video & Generative Video Systems

ONE2X founder frames video generation strategy around DSL, context and data. • Founder left a base-model team to start ONE2X • Medeo focuses on video DSL (atomic capabilities) • Architecture: DSL → Context → Environment • Compression equals intelligence principle emphasized • Data stages: public, domain, endogenous • Endogenous data is the startup moat target • System intelligence measured by token efficiency • Product must convert creation into expressive 'expression' • Large AI short-drama studios still use human labor • Seko 2.0 supports up to 100-episode long series • BulletTime research enables freeze-frame 3D camera moves • Video generation still needs model improvements for fragments
Key insights
ONE2X built Medeo focusing on DSL and context engineering
Founder left a base-model company to build ONE2X and Medeo, arguing product success depends on constructing a domain-specific language (DSL), context layer, and user environment rather than only on foundation models. The architecture splits into DSL (atomic video capabilities), Context (entropy reduction and precise instructions), and Environment (user interactions as explicit data), with emphasis on generating new 'inborn' data to form moats.
Compression-as-intelligence and data stages inform strategy
The founder described 'compression is intelligence' and three data stages: public data, domain/private data, and endogenous (product-generated) data. ONE2X targets endogenous data creation to build defensible assets and measure system intelligence by lower token cost to reach outcomes.
Video generation industry still needs human-in-loop for now
Reports note many AI short-drama studios still rely heavily on human labor; largest firms reportedly employ over 1,000 people manually producing AI short dramas. Seko 2.0 supports series up to 100 episodes, but current multi-modal models and video fragment generation still have issues requiring human fixes.
Research advances: BulletTime enables 3D camera moves from video
Stanford and ETH released BulletTime, a system that uses a single video to freeze or slow time while enabling arbitrary 3D camera movement; ten demonstration examples were published showing image-to-3D camera tricks.
Product and monetization views: focus on internal data and expert users
ONE2X favors building 'environments' that produce high-quality endogenous data and prioritize expert early users whose interactions raise system intelligence. The company measured early traction with an organic video achieving millions of views, validating product-market signals for their approach.