NVIDIA GeForce RTX — AI on your own computer
Three films for NVIDIA on generative AI on a PC with a GeForce RTX 5090: images and video in ComfyUI plus automation with n8n.
The brief
An influencer project for NVIDIA, commissioned by Gamma PR. The campaign set out to show AI models running locally on a graphics card. I wrote three scripts and produced the videos, recorded on a PC with an NVIDIA GeForce RTX 5090 and NVIDIA Studio drivers.
3 films on local AI in ComfyUI
Each film is a separate test on the same computer: images, video and automation.
1. Images in ComfyUI: NVFP4 vs BF16
NVIDIA optimised its drivers for ComfyUI: native support for the NVFP4 format brings up to three times the performance and 60% lower memory use. I tested it on the Z-Image template. In BF16 the first image, including model loading, took 11 seconds and each next one about 3.4 seconds. In NVFP4 the first took 3.7 seconds and each next one about 2 seconds, with the same image quality. That pace works great for storyboards and quick frame checks. You can also watch the full film on YouTube.
2. Video in ComfyUI: the LTX-2 model
I ran the open LTX-2 model from Lightricks from a ComfyUI template: a start frame, a prompt and that is it. A four-second 720p shot took 105 seconds including model loading, further variants under a minute, and a Full HD 1920 × 1080 version 104 seconds. That is enough to build an animatic, show it to the client and run a focus group before the actual production. More on local video generation in the NVIDIA guide. You can also watch the full film on YouTube.
3. Automation: n8n and ComfyUI
Local AI makes real business sense when there is no internet on site, data has to stay on the device, or the scale of a project outgrows the budget for cloud credits. I showed how to set up n8n on your own computer and connect it to ComfyUI: prompts and aspect ratios read from a Google Sheet, a loop over the rows, the Z-Image workflow exported as an API, finished files saved to Google Drive and a “done” status written back to the sheet. That is how I prepared over 400 product images in a single night. You can also watch the full film on YouTube.
Have a similar project?
Tell me what you are working on and I will get back to you with a proposal.