Understand this story
FastVideo has open-sourced Dreamverse Harness, an agentic video production framework that enables AI agents to collaborate throughout the workflow, from scriptwriting and storyboarding to final editing, with support for fully local operation or cloud services. Built on DeepSeek Harness, the framework allows models to be swapped and workflow steps to be added or removed. It can schedule model and shot tasks across up to four NVIDIA DGX Spark systems, and also supports conversational scene editing and upscaling selected shots to 2K. There are currently no further developments to report; support for Blender and ComfyUI is still under development.
AI-generated from public reports
Latest development · Report
FastVideo Open-Sources Dreamverse Harness, an AI Agent for Video ProductionDevelopments
1 confirmed developments
- Source:UCSD Hao AI Lab
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 20
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, an AI Agent for Video Production
FastVideo has open-sourced Dreamverse Harness, enabling AI agents to collaborate on the entire video production process, from scripting to final editing, and allowing it to run locally on NVIDIA DGX Spark or use cloud services. It supports custom models and workflows, with plans to add Blender and ComfyUI; a single device can host both the director agent and video model simultaneously, while the system can scale to coordinated operation across up to four devices. GitHub: https://t.co/Xcvadrksnd
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, an Agent Framework for Video Production
FastVideo has open-sourced Dreamverse Harness, enabling AI agents to collaborate on the entire video production process from scripting and storyboarding to editing, with the option to run entirely locally on NVIDIA DGX Spark or use cloud services. A single Spark can run both the director agent and video model simultaneously within 128 GB of unified memory. The system supports collaboration among up to four units, with two units taking approximately 158 seconds to re-render video at 2K resolution. Built on Deepseek Harness, the framework allows models to be swapped and workflow steps to be added or removed, with support for Blender and ComfyUI planned.
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources the Dreamverse Harness AI Agent Tool for Video Production
FastVideo has open-sourced Dreamverse Harness, enabling AI agents to collaborate on the entire video production process from scriptwriting to final editing, with support for either fully local operation or cloud services. Users can connect local or hosted models, add or remove workflow steps, and scale computing power across up to four NVIDIA DGX Sparks; the team is adding support for Blender and ComfyUI.
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, a Video Production Tool
FastVideo has open-sourced Dreamverse Harness, which enables AI agents to assist with the entire video production workflow, from script to final edit, and supports either fully local operation or cloud services. It can manage models and tasks across up to four NVIDIA DGX Spark systems, assigning shots to idle devices and switching models as needed. With four devices, two can re-render selected shots in 2K in about 158 seconds while the other two continue generating new shots.
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, a Video Production Agent
FastVideo has open-sourced Dreamverse Harness, enabling AI agents to assist with the entire video production workflow, from scripts and storyboards to editing and 2K upscaling. The system can run entirely locally: the 128 GB of unified memory on a single NVIDIA DGX Spark can simultaneously accommodate a director agent served by vLLM and a video model, and the system can scale to four devices. Built on Deepseek Harness, it supports swapping models and adding or removing workflow steps, with support for Blender and ComfyUI planned for the future.
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, Enabling Agents to Handle End-to-End Video Production
FastVideo has open-sourced Dreamverse Harness, enabling agents to collaborate on the entire video production process from scriptwriting to final editing, either locally on NVIDIA DGX Spark or using cloud services. It can write scripts and story beats, design camera positions and actions for each shot, modify scenes through conversation, and upscale selected clips to 2K. A single DGX Spark can simultaneously host the director agent served by vLLM and the video model within 128 GB of unified memory, and the system can scale to up to four devices.
UCSD Hao AI LabFirst-partySelected
FastVideo Open-Sources Dreamverse Harness, an Agent Framework for Video Production
FastVideo has open-sourced Dreamverse Harness, enabling AI agents to assist with the video production process from scripting and storyboarding through final editing, with support for fully local operation or cloud services. The framework can schedule models and shot-generation tasks across up to four NVIDIA DGX Spark systems, supports conversational scene editing, and can upscale selected shots to 2K. Built on DeepSeek Harness, it allows models to be swapped and workflow steps to be added or removed, while support for Blender and ComfyUI is still under development.
AI Trends