AI video generation has come to ComfyUI, where developers say Stable Video Diffusion can run on an Nvidia GTX 1080 with 8 gigabytes of VRAM. The reported example produces a 25-frame video at 1024 x 576 in about 3 minutes, bringing the research preview to hardware that is 7 years old.
ComfyUI brings Stable Video Diffusion to local workflows
ComfyUI is a graphical interface for Stable Diffusion that uses a graph and node layout. That format lets users connect components into complex workflows, and the new update adds support for Stability AI’s Stable Video Diffusion models.
This gives users another way to work with the generative video model. The source describes ComfyUI as an alternative to interfaces such as AUTOMATIC1111, while emphasizing the hardware requirements reported by the developers.
What the reported hardware can produce
On a 7-year-old Nvidia GTX 1080 equipped with 8 gigabytes of VRAM, the ComfyUI update can generate video at 1024 x 576 resolution with a length of 25 frames, according to its developers. They estimate that creating a video takes about 3 minutes.
AMD users can also run Stable Video Diffusion through ComfyUI. The reported setup uses an AMD 6800 XT with ROCm on Linux, extending the available hardware options beyond the Nvidia example.
These details describe particular configurations and an example output. They help set expectations for local AI video generation, but the source does not provide performance figures for other hardware or workflow settings.
Two frame lengths and sample workflows
Stable Video Diffusion is based on the Stable Diffusion image model and comes in two frame-to-video formats. One generates 14 frames and the other 25 frames; both support adjustable frame rates between 3 and 30 frames per second.
The ComfyUI developers published two sample workflows on their blog, one for each model length. Those examples give users a starting point for trying the 14-frame or 25-frame version in the node-based interface.
A research preview with more plans ahead
Stability AI released Stable Video Diffusion as a research preview earlier in the week. The model was designed to outperform commercial competitors RunwayML and Pika Labs in user preference studies, according to the source.
The initial release was available as a research version on Github. Stability AI said it planned to build an ecosystem of models around the technology, with a freely available final model and a web version with text-to-video functionality also planned.
For now, the ComfyUI update offers a practical route to explore the frame-to-video models locally. Its published examples show how a workflow can connect an image-based foundation to short video output, while future plans point to additional models and a text-to-video web experience.