Research Scientist, Long Video Generation Job at Hedra, Inc in San Francisco

About Hedra

Hedra is a pioneering generative media company backed by top investors at Index, A16Z, and Abstract Ventures. We're building Hedra Studio, a multimodal creation platform capable of control, emotion, and creative intelligence.

At the core of Hedra Studio is our Character-3 foundation model, the first omnimodal model in production. Character-3 jointly reasons across image, text, and audio for more intelligent video generation — it’s the next evolution of AI-driven content creation.

At Hedra, we’re a team of hard-working, passionate individuals seeking to fundamentally change content creation and build a generational company together. We value startup energy, initiative, and the ability to turn bold ideas into real products. Our team is fully in-person in SF / NY with a shared love for whiteboard problem-solving.

Overview

We are seeking a highly motivated Research Scientist to push the limits of long-form video generation, with a focus on auto-regressive modeling, causal attention mechanisms, and efficient sequence handling. The ideal candidate will have a deep understanding of temporal modeling in generative AI and experience building scalable architectures for multi-minute coherent video outputs.

Responsibilities

Design and implement long video generation architectures, with emphasis on auto-regressive generation, causal attention, and memory-efficient transformer designs.

Develop methods for maintaining temporal and semantic coherence over long time horizons.

Work closely with engineering to integrate research into production-grade pipelines.

Stay on top of recent advances in long-context transformers, sequence compression, and scalable video generation.

Present results internally and externally, including possible top-tier conference submissions.

Qualifications

PhD or strong research / industry experience in Computer Science, Machine Learning, or related fields, with a focus on sequence modeling or generative models.

Deep understanding of transformer architectures, attention mechanisms, and auto-regressive modeling.

Experience with long-context processing and memory-efficient computation.

Proficiency in Python and PyTorch; ability to rapidly prototype and iterate on new architectures.

A record of impactful research or large-scale system deployments.

Benefits

Competitive compensation + equity

401k (no match)

Healthcare (Silver PPO Medical, Vision, Dental)

Lunch and snacks at the office

We encourage you to apply even if you don't meet every requirement — we value curiosity, creativity, and the drive to solve hard problems.

#J-18808-Ljbffr

In Summary : Hedra is a pioneering generative media company backed by top investors at Index, A16Z, and Abstract Ventures . We're building Hedra Studio, a multimodal creation platform capable of control, emotion, and creative intelligence . The ideal candidate will have a deep understanding of temporal modeling in generative AI .

En Español : Hedra es una empresa de medios generativos pionera respaldada por los principales inversores de Index, A16Z y Abstract Ventures. Estamos construyendo Hedra Studio, una plataforma de creación multimodal capaz de control, emoción e inteligencia creativa. En el núcleo de Hedra Studios se encuentra nuestro modelo fundacional Character-3, el primer modelo omnimodal en producción. Trabajar estrechamente con la ingeniería para integrar la investigación en las tuberías de producción. Manténgase al tanto de los recientes avances en transformadores de contexto largo, compresión de secuencias y generación de video escalable. Presente resultados internamente y externamente, incluidas posibles presentaciones de conferencias de primer nivel. Calificaciones doctoradas o una sólida experiencia de investigación / industria en Ciencia informática, aprendizaje automático o campos relacionados, con un enfoque en el modelado de secuencia o modelos generativos. Una comprensión profunda de las arquitecturas de transformador, mecanismos de atención y modelo auto-regresivo. Experiencia en procesamiento de contextos largos y computación eficiente en memoria. Proficiencia en Python y PyTorch; capacidad para crear prototipos rápidamente e iterar sobre nuevas arquitectura.