23 Daily — According to the project's description, the open-source project 23daily/ComfyUI-Vast-Worker provides a solution for Mac users to run ComfyUI workflows via rented GPUs on Vast.ai. Users can run their existing workflows without modification, while custom nodes and models automatically synchronize and download, enabling Mac users to accelerate their ComfyUI executions at lower cost and barrier.
Features and Workflow Operation
The plugin was originally built and tested on macOS, targeting users whose local machines excel at editing workflows but are too slow for running them. The local ComfyUI remains the editor; by toggling a switch in the sidebar and pressing 'Run', the entire workflow is sent remotely to a Vast.ai GPU instance rented, set up, and destroyed on demand.
The Vast.ai instance automatically boots up, downloads required models from sources like Hugging Face or Civitai, runs the workflow, and sends results back to the local output folder. This requires no manual Docker, SSH, or remote environment configuration from the user.
Custom node packs used in the workflow are installed automatically on the rented machine at the same local version, and models not available locally are uploaded automatically.
Advantages Compared to Self-Managed GPU Workers
Unlike managing your own remote GPU workstation, this plugin requires no changes to workflows for remote execution, allowing users to run their current ComfyUI workflows directly.
Custom nodes and models automatically sync, removing the need for manual updating and transfer.
Hardware is rented hourly with flexible selection by GPU type and region, and boot time is a few minutes, providing flexibility compared to an always-on worker owned by the user.
This solution suits users who want to scale GPU usage beyond their local hardware without maintaining a second ComfyUI instance.
Usage Instructions and Workflow
Users paste their Vast.ai API key into the Vast Worker tab in the sidebar, configure price limits, GPU models, and preferred regions, which filters appropriate rental offers. Switching 'Run on' to Vast.ai and hitting Run allows selection and confirmation of the machine to rent and run remotely. Initial start takes a few minutes, subsequent runs reuse the rented instance.
Models are downloaded primarily on the remote instance; if unavailable for remote download but present locally, they upload automatically. Reference images and other local files synchronize to the remote instance.
The rented instance auto-destroys after idle timeout (default 15 minutes) or can be manually destroyed immediately.
Editor's take: Opportunities for Entrepreneurs and Validation Methods
The core value is breaking local GPU limitations by enabling creators and developers to flexibly and cost-effectively accelerate workflows via cloud GPUs. This is an attractive lightweight alternative for users facing high hardware costs or uneven workloads.
Entrepreneurs can leverage this to craft service-centric solutions combining flexible GPU rental, reducing customer entry barriers, and building scalable AI or graphics products. Validation can start by testing user experience, ensuring stable seamless remote run switching, cost advantages over self-hosted setups, and reliability of custom node and model synchronization.
Practical validation methods include running mid-to-large scale workflows remotely, measuring response times, compute quality, and system robustness.